Daily Archives: May 18, 2023

Stability AI is releasing an open-source version of DreamStudio, a commercial interface for the companys AI image generator model, Stable Diffusion. In a press statement on Wednesday, Stability AI said the new release dubbed StableStudio marks a fresh chapter for the platform and will serve as a showcase for the companys dedication to advancing open-source development.

Making an open-source version of DreamStudio carries benefits for Stability AI. It allows community developers to improve and experiment with the interface, with the company potentially reaping the rewards conferred by these improvements. Stability AI stressed community building in its press release, noting how from enabling local-first development, to experimenting with a new plugin system, weve tried hard to make things extensible for external developers.

Stability AIs approach to open-source development has helped drive interest in its products

Stability AI has previously leaned hard on its open-source approach to create interest in its products. Various versions of Stable Diffusion have been freely available to download and tinker with since the model was publicly released back in August 2022, and last month, the company released a suite of open-source large language models (LLMs) collectively called StableLM. Stability AIs founder and CEO, Emad Mostaque, has been outspoken about the importance of making AI tools open source in order to increase public trust, claiming that open models will be essential for private data, in a Zoom call with the press last month.

However, the companys approach sometimes seems to lack direction, too. For example, StableStudio will be available alongside DreamStudio and potentially compete with it. The company has previously said it plans to generate revenue by creating customized versions of DreamStudio for corporate clients, but its not clear how successful this strategy has been. Recent reports suggest the firm is burning through cash and note that its most important models, like Stable Diffusion, were built in collaboration with other parties.

Go here to read the rest:

Stability AI releases StableStudio in latest push for open-source AI - The Verge

Companies across industries are using artificial Intelligence (AI). And that has people thinking about what AI means for the future of many professions. Alphabet (GOOG 1.16%) (GOOGL 1.11%), the parent company of Google, is no stranger to AI. The company uses AI throughout its business -- from optimizing its search capabilities to better serving its advertising customers.

Recently, Alphabet chief executive officer Sundar Pichai spoke about how AI is set to revolutionize the working world. He even said one industry in particular might be transformed by the technology. And, surprisingly, Pichai says this may not result in layoffs -- but instead, more jobs.

Pichai told The Verge in an interview that AI may make the law profession better in certain ways and may result in more people actually becoming lawyers. Law firms today already use AI to help draft documents, verify contracts, and complete other tasks. The idea is AI won't take away lawyers' jobs. Instead, it will help them and their staff do certain things more quickly -- and give them more time to focus on more complicated parts of the job.

"I'm willing to almost bet 10 years from now, maybe there are more lawyers," Pichai said in the interview.

Pichai said that some of the jitters people have about AI today are a lot like the worries they had years ago with the introduction of new technology like computers or the Internet. Yes, these technologies may have hurt some jobs, but they've also introduced many new ones and resulted in overall progression in the employment market.

Pichai is doing his part to advance the use of AI. He's made it a focus at Alphabet. For instance, the company uses it to help the Google search engine better understand the meaning of each vocal and written search. Another example, this time in the area of advertising: An AI tool predicts how much advertisers should spend to meet their campaign goals.

So, the use of AI isn't necessarily about eliminating jobs. It's about improving how we do them. And this makes AI a great new technology to invest in today. Alphabet itself is a solid AI stock to buy -- and investors will like the company for other reasons too, such as its giant market share in the search market. Google Search holds 92% of that market. And its use of AI may keep that going.

You also can invest in AI by buying shares of e-commerce giant Amazon (AMZN 1.85%). The company has used AI for years -- that's what helps Amazon recommend products to you that you may like, for example. And last year, Amazoneven bought a company called Snackable.AI. Amazon intends to use the company's machine learning tools to boost its streaming and podcast capabilities.

But AI isn't limited to tech companies. You'll find companies in healthcare turning their attention to the area too. Vaccine maker Moderna (MRNA 0.37%) used AI to help it in the development of the coronavirus vaccine -- and is using it to make research and development more efficient. Moderna even recently signed a deal with International Business Machines to use that company's AI platform in the drug development process. And Medtronic is using AI in many ways, such as predicting outcomes in spine surgery.

That means there are plenty of stocks today offering you opportunities for AI investing. So, this technology may transform jobs -- as Pichai says -- and it may also open the door to new investment possibilities.

Suzanne Frey, an executive at Alphabet, is a member of The Motley Fool's board of directors. John Mackey, former CEO of Whole Foods Market, an Amazon subsidiary, is a member of The Motley Fool's board of directors. Adria Cimino has positions in Amazon.com. The Motley Fool has positions in and recommends Alphabet and Amazon.com. The Motley Fool recommends Moderna. The Motley Fool has a disclosure policy.

See original here:

Google CEO Sundar Pichai Predicts That This Profession Will Be ... - The Motley Fool

Frances privacy watchdog, the CNIL, has published an action plan for artificial intelligence which gives a snapshot of where it will be focusing its attention, including on generative AI technologies like OpenAIs ChatGPT, in the coming months and beyond.

A dedicated Artificial Intelligence Service has been set up within the CNIL to work on scoping the tech and producing recommendations for privacy-friendly AI systems.

A key stated goal for the regulator is to steer the development of AI that respects personal data, such as by developing the means to audit and control AI systems to protect people.

Understanding how AI systems impact people is another main focus, along with support for innovative players in the local AI ecosystem which apply the CNILs best practice.

The CNIL wants to establish clear rules protectingthe personal data of European citizens in order to contribute to the development of privacy-friendly AI systems, it writes.

Barely a week goes by without another bunch of high profile calls from technologists asking regulators to get to grips with AI. And just yesterday, during testimony in the US Senate, OpenAIs CEO Sam Altman called for lawmakers to regulate the technology, suggesting a licensing and testing regime.

However data protection regulators in Europe are far down the road already with the likes of Clearview AI already widely sanctioned across the bloc for misuse of peoples data, for example. While the AI chatbot, Replika, has faced recent enforcement in Italy.

OpenAIs ChatGPT also attracted a very public intervention by the Italian DPA at the end of March which led to the company rushing out with new disclosures and controls for users, letting them apply some limits on how it can use their information.

At the same time, EU lawmakers are in the process of hammering out agreement on a risk-based framework for regulating applications of AI which the bloc proposed back in April 2021.

This framework, the EU AI Act, could be adopted by the end of the year and the planned regulation is another reason the CNIL highlights for preparing its AI action plan, saying the work will also make it possible to prepare for the entry into application of the draft European AI Regulation, which is currently under discussion.

Existing data protection authorities (DPAs) are likely to play a role in enforcement of the AI Act so regulators building up AI understanding and expertise will be crucial for the regime to function effectively. While the topics and details EU DPAs choose focus their attention on are set to weight the operational parameters of AI in the future certainly in Europe and, potentially, further afield given how far ahead the bloc is when it comes to digital rule-making.

On generative AI, the French privacy regulator is paying special attention to the practice by certain AI model makers of scraping data off the Internet to build data-sets for training AI systems like large language models (LLMs) which can, for example, parse natural language and respond in a human-like way to communications.

It says a priority area for its AI service will be the protection of publicly available data on the web against the use of scraping, or scraping, of data for the design of tools.

This is an uncomfortable area for makers of LLMs like ChatGPT that have relied upon quietly scraping vast amounts of web data to repurpose as training fodder. Those that have hoovered up web information which contains personal data face a specific legal challenge in Europe where the General Data Protection Regulation (GDPR), in application since May 2018, requires them to have a legal basis for such processing.

There are a number of legal bases set out in the GDPR however possible options for a technology like ChatGPT are limited.

In the Italian DPAs view, there are just two possibilities: Consent or legitimate interests. And since OpenAI did not ask individual web users for their permission before ingesting their data the company is now relying on a claim of legitimate interests in Italy for the processing; a claim that remains under investigation by the local regulator, Garante. (Reminder: GDPR penalties can scale up to 4% of global annual turnover in addition to any corrective orders.)

The pan-EU regulation contains further requirements to entities processing personal data such as that the processing must be fair and transparent. So there are additional legal challenges for tools like ChatGPT to avoid falling foul of the law.

And notably in its action plan, Frances CNIL highlights the fairness and transparency of the data processing underlying the operation of [AI tools] as a particular question of interest that it says its Artificial Intelligence Service and another internal unit, the CNIL Digital Innovation Laboratory, will prioritize for scrutiny in the coming months.

Other stated priority areas the CNIL flags for its AI scoping are:

Giving testimony to a US senate committee yesterday, Altman was questioned by US lawmakers about the companys approach to protecting privacy and the OpenAI CEO sought to narrowly frame the topic as referring only to information actively provided by users of the AI chatbot noting, for example, that ChatGPT lets users specify they dont want their conversational history used as training data. (A feature it did not offer initially, however.)

Asked what specific steps its taken to protect privacy, Altman told the senate committee: We dont train on any data submitted to our API. So if youre a business customer of ours and submit data, we dont train on it at all If you use ChatGPT you can opt out of us training on your data. You can also delete your conversation history or your whole account.

But he had nothing to say about the data used to train the model in the first place.

Altmans narrow framing of what privacy means sidestepped the foundational question of the legality of training data. Call it the original privacy sin of generative AI, if you will. But its clear that eliding this topic is going to get increasingly difficult for OpenAI and its data-scraping ilk as regulators in Europe get on with enforcing the regions existing privacy laws on powerful AI systems.

In OpenAIs case, it will continue to be subject to a patchwork of enforcement approaches across Europe as it does not have an established base in the region which the GDPRs one-stop-shop mechanism does not apply (as it typically does for Big Tech) so any DPA is competent to regulate if it believes local users data is being processed and their rights are at risk.So while Italy went in hard earlier this year with an intervention on ChatGPT that imposed a stop-processing-order in parallel to it opening an investigation of the tool, Frances watchdog only announced an investigation back in April, in response to complaints. (Spain has also said its probing the tech, again without any additional actions as yet.)

In another difference between EU DPAs, the CNIL appears to be concerned about interrogating a wider array of issues than Italys preliminary list including considering how the GDPRs purpose limitation principle should apply to large language models like ChatGPT. Which suggests it could end up ordering a more expansive array of operational changes if it concludes the GDPR is being breached.

The CNIL will soon submit to a consultation a guide on the rules applicable to the sharing and re-use of data, it writes. This work will include the issue of re-use of freely accessible data on the internet and now used for learning many AI models. This guide will therefore be relevant for some of the data processing necessary for the design of AI systems, including generative AIs.

It will also continue its work on designing AI systems and building databases for machine learning. These will give riseto several publications starting in the summer of 2023, following the consultation which has already been organised with several actors, in order to provide concrete recommendations, in particular as regards the design of AI systems such as ChatGPT.

Heres the rest of the topics the CNIL says will be gradually addressed via future publications and AI guidance it produces:

On audit and control of AI systems, the French regulator stipulates that its actions this year will focus on three areas: Compliance with an existing position on the use of enhanced video surveillance, which it published in 2022; the use of AI to fight fraud (such as social insurance fraud); and on investigating complaints.

It also confirms it has already received complaints about the legal framework for the training and use of generative AIs and says its working on clarifications there.

The CNIL has, in particular, received several complaints against the company OpenAI which manages the ChatGPT service, and has opened a control procedure, it adds, noting the existence of a dedicated working group that was recently set up within the European Data Protection Board to try to coordinated how different European authorities approach regulating the AI chatbot (and produce what it bill as a harmonised analysis of the data processing implemented by the OpenAI tool).

In further words of warning for AI systems makers who never asked peoples permission to use their data, and may be hoping for future forgiveness, the CNIL notes that itll be paying particular attention to whether entities processing personal data to develop, train or use AI systems have:

As for support for innovative AI players that want to be compliant with European rules (and values), the CNIL has had a regulatory sandbox up and running for a couple of years and its encouraging AI companies and researchers working on developing AI systems that play nice with personal data protection rules to get in touch (via ia@cnil.fr).

Read more here:

Frances privacy watchdog eyes protection against data scraping in AI action plan - TechCrunch

Solving consumer engagement is worth $1 trillion to our organization, said a health plan executive to me one day. He was obviously being a bit sensationalist with the magnitude of the number, but its been a common thread in strategic dialogue across all of healthcare for quite a whilethat so much of the high cost, waste, and poor clinical outcomes from which our healthcare system suffers stems from the lack of individuals engagement with itor the flip, which is healthcare organizations inability to engage, at scale and cost-effectively, with its patients and members.

And its not just consumer engagement that needs to be solvedour countrys severe clinician burnout problem, combined with overall high rates of staff churn, are causing healthcare organizations to falter at delivering on their core services. One study shows that the average hospital has turned over 100.5% of its workforce in the last 5 years!

In this context, when it comes to generative AI, healthcare is an industry that we view as holding the most potential for tangible and measurable impact. Closing the gap on a shortage of millions of healthcare workers in the next several years, while also trying to increase leverage for those already in the workforce, requires much more than just the traditional paths of training or importing more human labor. Were excited to be backing Hippocratic AI as they apply generative AI to execute against this opportunity set.

Imagine a world in which every patient, provider, and administrative staff member could interact with an immediately available, fully context-aware, completely capable, and charismatic conversationalist to help each individual pick the right path or do their job better (a form of always-on triage, as weve described in the past). Imagine that the marginal cost of engaging a patient through empathetic phone calls was on the order of $0.10 per hour, as opposed to the $50+ it might cost today. The very nature of generative AIconversational, scalable, accessible to non-technical usershas the potential to solve the shortcomings of previous generations of rules-based chatbots and other such products in making these concepts a reality.

But AI applications in healthcare also pose among the highest stakes of any industry. AI skeptics might point to the lack of focus on responsibility, safety, and regulatory compliance exhibited by many companies in this space. Not to mention the challenge of assembling a cross-disciplinary team with deep expertise in LLM development, healthcare delivery, and healthcare administration to build AI products that actually work.

Hippocratic AIs name alone represents their safety-first ethos (referring to the Hippocratic Oath that physicians commit to, in which the core principles are to do no harm to patients and to maintain confidentiality of a patients medical information). Theyve built a unique framework to incorporate professional-grade certification, RLHF (reinforcement learning from human feedback) through a panel of healthcare professionals, and bedside manner into their non-diagnostic, patient-facing conversational LLMs, with the recognition that passing a medical board exam is not enough to ensure that a model is ready to be deployed into a real-world setting.

Weve known the CEO, Munjal Shah, since investing in his last company in 2017 (which was his third, after previously selling an AI company to Google), and thus know he has uniquely earned secrets about how to build a company at the intersection of AI and healthcare. He most recently ran a Medicare brokerage business that involved a national-scale call center that made personalized recommendations to seniors based on their individual claims history. There, he led through the operational pains of scaling an empathetic but efficient engagement platform for consumers in a regulated healthcare context. We believe these competencies give him and his founding team (composed of individuals with clinical, LLM development, and healthcare operations experience) an edge in understanding what it takes to bring high-impact, responsible, and safe generative AI products to market, and consider it a privilege to be backing him again.

***

The views expressed here are those of the individual AH Capital Management, L.L.C. (a16z) personnel quoted and are not the views of a16z or its affiliates. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, this content may include third-party advertisements; a16z has not reviewed such advertisements and does not endorse any advertising content contained therein.

This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by Andreessen Horowitz (excluding investments for which the issuer has not provided permission for a16z to disclose publicly as well as unannounced investments in publicly traded digital assets) is available at https://a16z.com/investments/.

Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Past performance is not indicative of future results. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see https://a16z.com/disclosures for additional important information.

See original here:

Investing in Hippocratic AI - Andreessen Horowitz

Published: May 17, 2023 at 9:54 a.m. ET

A few months back, some on Wall Street were worried that Microsoft Corp.s artificial-intelligence advancements would help the company eat into Alphabet Inc.s dominant Google search business.

Those concerns may be soon fading, as the investment community has rightfully determined that Google is in good shape around AI, Barclays analyst Ross Sandler wrote in a note to clients Wednesday. But the search giant still faces other challenges as generative AI, the type of artificial intelligence popularized by ChatGPT, becomes...

A few months back, some on Wall Street were worried that Microsoft Corp.s artificial-intelligence advancements would help the company eat into Alphabet Inc.s dominant Google search business.

Those concerns may be soon fading, as the investment community has rightfully determined that Google is in good shape around AI, Barclays analyst Ross Sandler wrote in a note to clients Wednesday. But the search giant still faces other challenges as generative AI, the type of artificial intelligence popularized by ChatGPT, becomes more prominent.

See also: Google developers conference is all about AI

The new elephant in the room, in Sandlers view, is whether Alphabets GOOGL GOOG own efforts with AI end up hurting its big moneymaker, at least in the short term. The concern is that as Google integrates generative AI into its search business, people might start to get more information that way, rather than clicking through to revenue-generating sources like sponsored listings.

Its fairly well documented that the top position on legacy SERPs [search engine results pages] gets the lions share of the clicks, anywhere from 2% to 39% depending on whether the first link is an ad or an organic result, Sandler wrote.

So the obvious problem for Google is that mobile and desktop search engines powered by generative AI are a complete departure from the typical look of search, he continued. Not only is the percentage of screen real estate dedicated to ads much lower, the entire layout and flow of the SERP appears to be changing in many cases.

Read: Alphabet stock proves popular with big funds

He gave an example of a mobile search for house plants thats dominated by ads in its traditional form but devoid of any ads above the fold in the demo generative-AI search layout.

The point being that any movement of commercial query share toward less commercial share using GenAI, would have air-pocket impacts, Sandler wrote. This speaks to just how good a business model traditional search really is in many cases, the ads are just as relevant if not more relevant than the organic results.

On the bright side, Google has a history of finding new ways to monetize search, and Sandler thinks its possible that more relevant AI-driven answers could ultimately help click-through rates for top positions on the page.

An auction with a winner of one could actually monetize at or above the traditional 10-blue-link SERP, he wrote. But this is going to be a big unknown for the foreseeable future.

Read the rest here:

As Alphabet flexes its AI prowess, there's a 'new elephant in the room' for Google - MarketWatch

This week, at its annual I/O developer conference in Mountain View, Google showcased ahead-spinning number of projects and products powered by or enhanced by AI. They included a new-and-improved version of itschatbot Bard, tools to help you write emails and documents or manipulate images,devices with AI baked in, and achatbot-like experimental version of Google search. For a full recap of the event, complete with insightful and witty commentary from my WIRED colleagues, check outour Google I/O liveblog.

Googles big pivot is, of course, largely fueled not by algorithms but by generative AI FOMO.

The appearance last November of ChatGPTthe remarkably clever butstill rather flawed chatbot fromOpenAIcombined with Microsoftadding the technology to its search engine Bing a few months later, triggered something of a panic at Google. ChatGPT proved wildly popular with users, demonstrating new ways to serve up information that threatened Googles vice grip on the search business and its reputation as the leader in AI.

The capabilities of ChatGPT and AI language algorithms like those powering itare so striking that some experts, including Geoffrey Hinton, a pioneering researcher who recently left Google, have felt compelled to warn that we might be building systems thatwe will someday struggle to control. OpenAIs chatbot is often astonishingly good at generating coherent text on a given subject, summarizing information from the web, and even answering extremely tricky questions that require expert knowledge.

And yet, unfettered AI language models are also silver-tongued agents of chaos. They will gladly fabricate facts, express unpleasant biases, andsay unpleasant or disturbing things with the right prompting. Microsoft was forced to limit the capabilities of Bing chat shortly after launch to avoid such embarrassing misbehavior, in part because its bot divulged its secret codenameSydneyandaccused aNew York Times columnist of not loving his spouse.

Google worked hard to tone down the chaotic streak of text-generation technology as it prepared theexperimental search feature announced yesterday that responds to search queries with chat-style answers synthesizing information from across the web.

Googles smarter version of search is impressively narrow-minded, refusing to use the first person or talk about its thoughts or feelings. It completely avoids topics that might be considered risky, refusing to dispense medical advice or offer answers on potentially controversial topics such as US politics.

Read the rest here:

The Boring Future of Generative AI | WIRED - WIRED

May 15 (Reuters) - OpenAI is preparing to release a new open-source language model to the public, The Information reported on Monday, citing a person with knowledge of the plan.

OpenAI's ChatGPT, known for producing prose or poetry on command, has gained widespread attention in Silicon Valley as investors see generative AI as the next big growth area for tech companies.

In January, Microsoft Corp (MSFT.O) announced a multi-billion dollar investment in OpenAI, deepening its ties with the startup and setting the stage for more competition with rival Alphabet Inc's (GOOGL.O) Google.

Meta Platforms Inc (META.O) is now rushing to join competitors Microsoft and Google in releasing generative AI products capable of creating human-like writing, art and other content.

OpenAI is unlikely to release a model that is competitive with GPT, the report said.

The company did not immediately respond to Reuters' request for a comment.

Reporting by Ananya Mariam Rajesh in Bengaluru; Editing by Shinjini Ganguli

Our Standards: The Thomson Reuters Trust Principles.

Link:

OpenAI readies new open-source AI model, The Information reports - Reuters.com

Amid the excitement surrounding generative AI since the release of ChatGPT, Bard, Claude, Midjourney, and other content-creating tools, CEOs are understandably wondering: Is this tech hype, or a game-changing opportunity? And if it is the latter, what is the value to my business?

The public-facing version of ChatGPT reached 100 million users in just two months. It democratized AI in a manner not previously seen while becoming by far the fastest-growing app ever. Its out-of-the-box accessibility makes generative AI different from all AI that came before it. Users dont need a degree in machine learning to interact with or derive value from it; nearly anyone who can ask questions can use it. And, as with other breakthrough technologies such as the personal computer or iPhone, one generative AI platform can give rise to many applications for audiences of any age or education level and in any location with internet access.

All of this is possible because generative AI chatbots are powered by foundation models, which are expansive neural networks trained on vast quantities of unstructured, unlabeled data in a variety of formats, such as text and audio. Foundation models can be used for a wide range of tasks. In contrast, previous generations of AI models were often narrow, meaning they could perform just one task, such as predicting customer churn. One foundation model, for example, can create an executive summary for a 20,000-word technical report on quantum computing, draft a go-to-market strategy for a tree-trimming business, and provide five different recipes for the ten ingredients in someones refrigerator. The downside to such versatility is that, for now, generative AI can sometimes provide less accurate results, placing renewed attention on AI risk management.

With proper guardrails in place, generative AI can not only unlock novel use cases for businesses but also speed up, scale, or otherwise improve existing ones. Imagine a customer sales call, for example. A specially trained AI model could suggest upselling opportunities to a salesperson, but until now those were usually based only on static customer data obtained before the start of the call, such as demographics and purchasing patterns. A generative AI tool might suggest upselling opportunities to the salesperson in real time based on the actual content of the conversation, drawing from internal customer data, external market trends, and social media influencer data. At the same time, generative AI could offer a first draft of a sales pitch for the salesperson to adapt and personalize.

The preceding example demonstrates the implications of the technology on one job role. But nearly every knowledge worker can likely benefit from teaming up with generative AI. In fact, while generative AI may eventually be used to automate some tasks, much of its value could derive from how software vendors embed the technology into everyday tools (for example, email or word-processing software) used by knowledge workers. Such upgraded tools could substantially increase productivity.

CEOs want to know if they should act nowand, if so, how to start. Some may see an opportunity to leapfrog the competition by reimagining how humans get work done with generative AI applications at their side. Others may want to exercise caution, experimenting with a few use cases and learning more before making any large investments. Companies will also have to assess whether they have the necessary technical expertise, technology and data architecture, operating model, and risk management processes that some of the more transformative implementations of generative AI will require.

The goal of this article is to help CEOs and their teams reflect on the value creation case for generative AI and how to start their journey. First, we offer a generative AI primer to help executives better understand the fast-evolving state of AI and the technical options available. The next section looks at how companies can participate in generative AI through four example cases targeted toward improving organizational effectiveness. These cases reflect what we are seeing among early adopters and shed light on the array of options across the technology, cost, and operating model requirements. Finally, we address the CEOs vital role in positioning an organization for success with generative AI.

Excitement around generative AI is palpable, and C-suite executives rightfully want to move ahead with thoughtful and intentional speed. We hope this article offers business leaders a balanced introduction into the promising world of generative AI.

Generative AI technology is advancing quickly (Exhibit 1). The release cycle, number of start-ups, and rapid integration into existing software applications are remarkable. In this section, we will discuss the breadth of generative AI applications and provide a brief explanation of the technology, including how it differs from traditional AI.

Generative AI can be used to automate, augment, and accelerate work. For the purposes of this article, we focus on ways generative AI can enhance work rather than on how it can replace the role of humans.

While text-generating chatbots such as ChatGPT have been receiving outsize attention, generative AI can enable capabilities across a broad range of content, including images, video, audio, and computer code. And it can perform several functions in organizations, including classifying, editing, summarizing, answering questions, and drafting new content. Each of these actions has the potential to create value by changing how work gets done at the activity level across business functions and workflows. Following are some examples.

As the technology evolves and matures, these kinds of generative AI can be increasingly integrated into enterprise workflows to automate tasks and directly perform specific actions (for example, automatically sending summary notes at the end of meetings). We already see tools emerging in this area.

As the name suggests, the primary way in which generative AI differs from previous forms of AI or analytics is that it can generate new content, often in unstructured forms (for example, written text or images) that arent naturally represented in tables with rows and columns (see sidebar Glossary for a list of terms associated with generative AI).

The underlying technology that enables generative AI to work is a class of artificial neural networks called foundation models. Artificial neural networks are inspired by the billions of neurons that are connected in the human brain. They are trained using deep learning, a term that alludes to the many (deep) layers within neural networks. Deep learning has powered many of the recent advances in AI.

However, some characteristics set foundation models apart from previous generations of deep learning models. To start, they can be trained on extremely large and varied sets of unstructured data. For example, a type of foundation model called a large language model can be trained on vast amounts of text that is publicly available on the internet and covers many different topics. While other deep learning models can operate on sizable amounts of unstructured data, they are usually trained on a more specific data set. For example, a model might be trained on a specific set of images to enable it to recognize certain objects in photographs.

In fact, other deep learning models often can perform only one such task. They can, for example, either classify objects in a photo or perform another function such as making a prediction. In contrast, one foundation model can perform both of these functions and generate content as well. Foundation models amass these capabilities by learning patterns and relationships from the broad training data they ingest, which, for example, enables them to predict the next word in a sentence. Thats how ChatGPT can answer questions about varied topics and how DALLE 2 and Stable Diffusion can produce images based on a description.

Given the versatility of a foundation model, companies can use the same one to implement multiple business use cases, something rarely achieved using earlier deep learning models. A foundation model that has incorporated information about a companys products could potentially be used both for answering customers questions and for supporting engineers in developing updated versions of the products. As a result, companies can stand up applications and realize their benefits much faster.

However, because of the way current foundation models work, they arent naturally suited to all applications. For example, large language models can be prone to hallucination, or answering questions with plausible but untrue assertions (see sidebar Using generative AI responsibly). Additionally, the underlying reasoning or sources for a response are not always provided. This means companies should be careful of integrating generative AI without human oversight in applications where errors can cause harm or where explainability is needed. Generative AI is also currently unsuited for directly analyzing large amounts of tabular data or solving advanced numerical-optimization problems. Researchers are working hard to address these limitations.

While foundation models serve as the brain of generative AI, an entire value chain is emerging to support the training and use of this technology (Exhibit 2). Specialized hardware provides the extensive compute power needed to train the models. Cloud platforms offer the ability to tap this hardware. MLOps and model hub providers offer the tools, technologies, and practices an organization needs to adapt a foundation model and deploy it within its end-user applications. Many companies are entering the market to offer applications built on top of foundation models that enable them to perform a specific task, such as helping a companys customers with service issues.

The first foundation models required high levels of investment to develop, given the substantial computational resources required to train them and the human effort required to refine them. As a result, they were developed primarily by a few tech giants, start-ups backed by significant investment, and some open-source research collectives (for example, BigScience). However, work is under way on both smaller models that can deliver effective results for some tasks and training thats more efficient. This could eventually open the market to more entrants. Some start-ups have already succeeded in developing their own modelsfor example, Cohere, Anthropic, and AI21 Labs build and train their own large language models.

CEOs should consider exploration of generative AI a must, not a maybe. Generative AI can create value in a wide range of use cases. The economics and technical requirements to start are not prohibitive, while the downside of inaction could be quickly falling behind competitors. Each CEO should work with the executive team to reflect on where and how to play. Some CEOs may decide that generative AI presents a transformative opportunity for their companies, offering a chance to reimagine everything from research and development to marketing and sales to customer operations. Others may choose to start small and scale later. Once the decision is made, there are technical pathways that AI experts can follow to execute the strategy, depending on the use case.

Much of the use (although not necessarily all of the value) from generative AI in an organization will come from workers employing features embedded in the software they already have. Email systems will provide an option to write the first drafts of messages. Productivity applications will create the first draft of a presentation based on a description. Financial software will generate a prose description of the notable features in a financial report. Customer-relationship-management systems will suggest ways to interact with customers. These features could accelerate the productivity of every knowledge worker.

But generative AI can also be more transformative in certain use cases. Following, we look at four examples of how companies in different industries are using generative AI today to reshape how work is done within their organization. The examples range from those requiring minimal resources to resource-intensive undertakings. (For a quick comparison of these examples and more technical detail, see Exhibit 3.)

The use cases outlined here offer powerful takeaways for CEOs as they embark on the generative AI journey:

The CEO has a crucial role to play in catalyzing a companys focus on generative AI. In this closing section, we discuss strategies that CEOs will want to keep in mind as they begin their journey. Many of them echo the responses of senior executives to previous waves of new technology. However, generative AI presents its own challenges, including managing a technology moving at a speed not seen in previous technology transitions.

Many organizations began exploring the possibilities for traditional AI through siloed experiments. Generative AI requires a more deliberate and coordinated approach given its unique risk considerations and the ability of foundation models to underpin multiple use cases across an organization. For example, a model fine-tuned using proprietary material to reflect the enterprises brand identity could be deployed across several use cases (for example, generating personalized marketing campaigns and product descriptions) and business functions, such as product development and marketing.

To that end, we recommend convening a cross-functional group of the companys leaders (for example, representing data science, engineering, legal, cybersecurity, marketing, design, and other business functions). Such a group can not only help identify and prioritize the highest-value use cases but also enable coordinated and safe implementation across the organization.

Generative AI is a powerful tool that can transform how organizations operate, with particular impact in certain business domains within the value chain (for example, marketing for a retailer or operations for a manufacturer). The ease of deploying generative AI can tempt organizations to apply it to sporadic use cases across the business. It is important to have a perspective on the family of use cases by domain that will have the most transformative potential across business functions. Organizations are reimagining the target state enabled by generative AI working in sync with other traditional AI applications, along with new ways of working that may not have been possible before.

A modern data and tech stack is key to nearly any successful approach to generative AI. CEOs should look to their chief technology officers to determine whether the company has the required technical capabilities in terms of computing resources, data systems, tools, and access to models (open source via model hubs or commercial via APIs).

For example, the lifeblood of generative AI is fluid access to data honed for a specific business context or problem. Companies that have not yet found ways to effectively harmonize and provide ready access to their data will be unable to fine-tune generative AI to unlock more of its potentially transformative uses. Equally important is to design a scalable data architecture that includes data governance and security procedures. Depending on the use case, the existing computing and tooling infrastructure (which can be sourced via a cloud provider or set up in-house) might also need upgrading. A clear data and infrastructure strategy anchored on the business value and competitive advantage derived from generative AI will be critical.

CEOs will want to avoid getting stuck in the planning stages. New models and applications are being developed and released rapidly. GPT-4, for example, was released in March 2023, following the release of ChatGPT (GPT-3.5) in November 2022 and GPT-3 in 2020. In the world of business, time is of the essence, and the fast-paced nature of generative AI technology demands that companies move quickly to take advantage of it. There are a few ways executives can keep moving at a steady clip.

Although generative AI is still in the early days, its important to showcase internally how it can affect a companys operating model, perhaps through a lighthouse approach. For example, one way forward is building a virtual expert that enables frontline workers to tap proprietary sources of knowledge and offer the most relevant content to customers. This has the potential to increase productivity, create enthusiasm, and enable an organization to test generative AI internally before scaling to customer-facing applications.

As with other waves of technical innovation, there will be proof-of-concept fatigue and many examples of companies stuck in pilot purgatory. But encouraging a proof of concept is still the best way to quickly test and refine a valuable business case before scaling to adjacent use cases. By focusing on early wins that deliver meaningful results, companies can build momentum and then scale out and up, leveraging the multipurpose nature of generative AI. This approach could enable companies to promote broader AI adoption and create the culture of innovation that is essential to maintaining a competitive edge. As outlined above, the cross-functional leadership team will want to make sure such proofs of concept are deliberate and coordinated.

As our four detailed use cases demonstrate, business leaders must balance value creation opportunities with the risks involved in generative AI. According to our recent Global AI Survey, most organizations dont mitigate most of the risks associated with traditional AI, even though more than half of organizations have already adopted the technology. Generative AI brings renewed attention to many of these same risks, such as the potential to perpetuate bias hidden in training data, while presenting new ones, such as its propensity to hallucinate.

As a result, the cross-functional leadership team will want to not only establish overarching ethical principles and guidelines for generative AI use but also develop a thorough understanding of the risks presented by each potential use case. It will be important to look for initial use cases that both align with the organizations overall risk tolerance and have structures in place to mitigate consequential risk. For example, a retail organization might prioritize a use case that has slightly lower value but also lower risksuch as creating initial drafts of marketing content and other tasks that keep a human in the loop. At the same time, the company might set aside a higher-value, high-risk use case such as a tool that automatically drafts and sends hyperpersonalized marketing emails. Such risk-forward practices can enable organizations to establish the controls necessary to properly manage generative AI and maintain compliance.

CEOs and their teams will also want to stay current with the latest developments in generative AI regulation, including rules related to consumer data protection and intellectual property rights, to protect the company from liability issues. Countries may take varying approaches to regulation, as they often already do with AI and data. Organizations may need to adapt their working approach to calibrate process management, culture, and talent management in a way that ensures they can handle the rapidly evolving regulatory environment and risks of generative AI at scale.

Business leaders should focus on building and maintaining a balanced set of alliances. A companys acquisitions and alliances strategy should continue to concentrate on building an ecosystem of partners tuned to different contexts and addressing what generative AI requires at all levels of the tech stack, while being careful to prevent vendor lock-in.

Partnering with the right companies can help accelerate execution. Organizations do not have to build out all applications or foundation models themselves. Instead, they can partner with generative AI vendors and experts to move more quickly. For instance, they can team up with model providers to customize models for a specific sector, or partner with infrastructure providers that offer support capabilities such as scalable cloud computing.

Companies can use the expertise of others and move quickly to take advantage of the latest generative AI technology. But generative AI models are just the tip of the spear: multiple additional elements are required for value creation.

To effectively apply generative AI for business value, companies need to build their technical capabilities and upskill their current workforce. This requires a concerted effort by leadership to identify the required capabilities based on the companys prioritized use cases, which will likely extend beyond technical roles to include a talent mix across engineering, data, design, risk, product, and other business functions.

As demonstrated in the use cases highlighted above, technical and talent needs vary widely depending on the nature of a given implementationfrom using off-the-shelf solutions to building a foundation model from scratch. For example, to build a generative model, a company may need PhD-level machine learning experts; on the other hand, to develop generative AI tools using existing models and SaaS offerings, a data engineer and a software engineer may be sufficient to lead the effort.

In addition to hiring the right talent, companies will want to train and educate their existing workforces. Prompt-based conversational user interfaces can make generative AI applications easy to use. But users still need to optimize their prompts, understand the technologys limitations, and know where and when they can acceptably integrate the application into their workflows. Leadership should provide clear guidelines on the use of generative AI tools and offer ongoing education and training to keep employees apprised of their risks. Fostering a culture of self-driven research and experimentation can also encourage employees to innovate processes and products that effectively incorporate these tools.

Businesses have been pursuing AI ambitions for years, and many have realized new revenue streams, product improvements, and operational efficiencies. Much of the successes in these areas have stemmed from AI technologies that remain the best tool for a particular job, and businesses should continue scaling such efforts. However, generative AI represents another promising leap forward and a world of new possibilities. While the technologys operational and risk scaffolding is still being built, business leaders know they should embark on the generative AI journey. But where and how should they start? The answer will vary from company to company as well as within an organization. Some will start big; others may undertake smaller experiments. The best approach will depend on a companys aspiration and risk appetite. Whatever the ambition, the key is to get under way and learn by doing.

Read this article:

What every CEO should know about generative AI - McKinsey

Wednesday 17 May 2023 10:55, UK

Artificial intelligence has produced its idea of what the "ideal" man and woman look like, based on social media data and results on the World Wide Web.

The AI images of men and women were created through engagement analytics on social media, using tools to look at billions of images of people.

The Bulimia Project, an eating disorder awareness group, monitored the findings and warned the results are "largely unrealistic" in their depiction of body types.

This is a limited version of the story so unfortunately this content is not available. Open the full version

It said the images of women tended to have a bias toward blonde hair, brown eyes and olive skin - while for men, there was a bias toward brown hair, brown eyes and olive skin.

It also found that AI's collection of social media-inspired images were "far more sexually charged" than those based on everything else it found on the World Wide Web.

The study also showed there was some variation between body preferences for men and women.

The images generated of the "perfect" female body according to social media in 2023 featured tanned and Caucasian-looking women with slim figures and small waists.

For women, 37% of the AI-generated images included blonde hair, while 53% of the images included women with olive skin.

Images of the "perfect" male body featured muscly men with a six-pack, wearing tight t-shirts.

The images were created using the AI image generators Dall-E 2, Stable Diffusion, and Midjourney.

For men, 67% of the AI-generated images included brown hair and 63% of the images included olive skin.

The Bulimia Project then asked AI to share its perspective based on images from across the internet.

Read more:Senator's chilling warning after AI imitates himNew iPhone feature can create a voice that sounds like youFirst human trial of dirty bomb antidote begins

For the "perfect" woman in 2023 - AI generated images of women mainly with brown eyes, brown hair and tanned skin. For men with the same prompt, it produced images of men with facial hair, predominantly with brown eyes and hair.

The Bulimia Project said: "Considering that social media uses algorithms based on which content gets the most lingering eyes, it's easy to guess why AI's renderings would come out more sexualised.

"But we can only assume that the reason AI came up with so many oddly shaped versions of the physiques it found on social media is that these platforms promote unrealistic body types, to begin with."

More:

AI creates images of the 'perfect' man and woman - Sky News

Search tools assisted by large language models (LLMs) are changing how researchers find scholarly information. One tool, scite Assistant, uses GPT-3.5 to generate answers from a database of millions of scientific papers. Another, Elicit, uses an LLM to write its answers to searches for articles in a scholarly database. Consensus finds and synthesizes research claims in papers, whereas SciSpace bills itself as an AI research assistant that can explain mathematics or text contained in scientific papers. All of these tools give natural-language answers to natural-language queries.

Search tools tailored to academic databases can use LLMs to offer alternative ways of identifying, ranking and accessing papers. In addition, researchers can use general artificial intelligence (AI)-assisted search systems, such as Bing, with queries that target only academic databases such as CORE, PubMed and Crossref.

All search systems affect scientists access to knowledge and influence how research is done. All have unique capabilities and limitations. Im intimately familiar with this from my experience building Search Smart, a tool that allows researchers to compare the capabilities of 93 conventional search tools, including Google Scholar and PubMed. AI-assisted, natural-language search tools will undoubtedly have an impact on research. The question is: how?

The time remaining before LLMs mass adoption in academic search must be used to understand the opportunities and limitations. Independent audits of these tools are crucial to ensure the future of knowledge access.

Tools such as ChatGPT threaten transparent science; here are our ground rules for their use

All search tools assisted by LLMs have limitations. LLMs can hallucinate: making up papers that dont exist, or summarizing content inaccurately by making up facts. Although dedicated academic LLM-assisted search systems are less likely to hallucinate because they are querying a set scientific database, the extent of their limitations is still unclear. And because AI-assisted search systems, even open-source ones, are black boxes their mechanisms for matching terms, ranking results and answering queries arent transparent methodical analysis is needed to learn whether they miss important results or systematically favour specific types of papers, for example. Anecdotally, I have found that Bing, scite Assistant and SciSpace tend to yield different results when a search is repeated, leading to irreproducibility. The lack of transparency means there are probably many limitations still to be found.

Already, Twitter threads and viral YouTube videos promise that AI-assisted search can speed up systematic reviews or facilitate brainstorming and knowledge summarization. If researchers are not aware of the limitations and biases of such systems, then research outcomes will deteriorate.

Regulations exist for LLMs in general, some within the sphere of the research community. For example, publishers and universities have hammered out policies to prevent LLM-enabled research misconduct such as misattribution, plagiarism or faking peer review. Institutions such as the US Food and Drug Administration rate and approve AIs for specific uses, and the European Commission is proposing its own legal framework on AI. But more-focused policies are needed specifically for LLM-assisted search.

Why open-source generative AI models are an ethical way forward for science

In working on Search Smart, I developed a way to assess the functionalities of databases and their search systems systematically and transparently. I often found capabilities or limitations that were omitted or inaccurately described in the search tools own frequently asked questions. At the time of our study, Google Scholar was researchers most widely used search engine. But we found that its ability to interpret Boolean search queries, such as ones involving OR and AND, was both inadequate and inadequately reported. On the basis of these findings, we recommended not relying on Google Scholar for the main search tasks in systematic reviews and meta-analyses (M. Gusenbauer & N. R. Haddaway Res. Synth. Methods 11, 181217; 2020).

Even if search AIs are black boxes, their performance can still be evaluated using metamorphic testing. This is a bit like a car-crash test: it asks only whether and how passengers survive varying crash scenarios, without needing to know how the car works internally. Similarly, AI testing should prioritize assessing performance in specific tasks.

LLM creators should not be relied on to do these tests. Instead, third parties should conduct a systematic audit of these systems functionalities. Organizations that already synthesize evidence and advocate for evidence-based practices, such as Cochrane or the Campbell Collaboration, would be ideal candidates. They could conduct audits themselves or jointly with other entities. Third-party auditors might want to partner with librarians, who are likely to have an important role in teaching information literacy around AI-assisted search.

The aim of these independent audits would not be to decide whether or not LLMs should be used, but to offer clear, practical guidelines so that AI-assisted searches are used only for tasks of which they are capable. For example, an audit might find that a tool can be used for searches that help to define the scope of a project, but cant reliably identify papers on the topic because of hallucination.

AI-assisted search systems must be tested before researchers inadvertently introduce biased results on a large scale. A clear understanding of what these systems can and cannot do can only improve scientific rigour.

M.G. is the founder of Smart Search, a free website that tests academic search systems.

Continued here:

Audit AI search tools now, before they skew research - Nature.com