OpenAI's o1 Accidentally Released

0:00Today on the AI Daily Brief, OpenAI's O1 model gets accidentally released, plus many other stories. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:21Hello, friends. Quick note, today is actually my wedding anniversary, and so this is going to be a slightly abbreviated episode. Instead of our normal headlines into main episode, we will be doing an extended headlines episode. Normal formats will be coming back for later in the week. But for now, let's dive in with reports on x slash Twitter that suggest that the full version of OpenAI's O1 model was briefly available on Friday night. According to the logs, the model was described as O1, our most capable model, great for tasks that require creativity and advanced reasoning. The model had an extremely long 200 ,000 token context window, which is twice as much as O1 Mini and almost four times the size of O1 Preview.

1:00Security researcher Anane Arora wrote, ChatGPT's O1 model is quote-unquote secretly accessible, even though the dropdown doesn't allow it. Image understanding works and the inference is incredibly fast. Surat Adesumali wrote, Tried to get O1 to solve a Sudoku. It couldn't solve it, but the thought process is still very impressive. I'm quite sure we'll have a model in a year or so which we'll be able to solve it. He continued, Sudoku is a great test of deductive reasoning. You can't solve it without thinking about it step-by-step. It's probably impossible to solve it any other way because of computational irreducibility.

1:31Sam Altman, unsurprisingly, was a little coy. He had previously tweeted, chat GPT growth is wild recently, very proud of the team. When following these reports, people started to ask for Sam to, quote, unleash full O1, he responded, not that much longer, hopefully. How long until the calls to unleash O2? I feel like when it comes to open AI, and particularly when it comes to leaks, you always have to wonder whether they were actually leaked or whether it was all part of a plan to build some hype. In any case, we shift now to enterprise AI usage, where Visa is going all in on generative AI. The company has deployed hundreds of AI use cases and is looking to deploy hundreds more.

2:07President of technology Rajat Tanija said, the company already has over 500 gen AI applications in use, the result of a go-fast strategy designed to reap the benefits of AI and keep up with increasingly sophisticated fraudsters. He said, this is a time when I think we have to innovate very fast. The news also comes amidst a global restructuring, which has seen 1 ,400 layoffs, including over 1 ,000 technology workers. A Visa spokesperson insisted, We don't invest in AI to displace our talent. We invest in AI to help our employees become more productive, continue to protect consumers from fraud, and to drive consistent innovation and payments.

2:38Visa gave examples of their Gen.AI uses, including a bug hunting tool and chatbots that are designed to be subject matter experts in a particular segment of the business. The company has also rolled out more pedestrian use cases like customer service bots. Overall, Visa say they have invested$3.3 billion in AI and data infrastructure over the past 10 years. Writes the Wall Street Journal, Visa's generative AI approach has pluses and minuses, said Gardner AI analyst Whit Andrews, adding that first movers quickly discover what works and what doesn't. Those who wait will learn from the mistakes of others and be able to just tackle the projects with the highest probability of success.

3:07More than 60 % of organizations are still in that second bucket, but it isn't clear yet which approach will pay more returns in the end. Now, that is a very quaint thing for a Gardner AI analyst to be able to say, given that their job is not on the line for not adopting AI fast enough. I don't believe that right now there is a single corporate officer in America who could reasonably and realistically suggest that their company just wait and see how everyone else figures it out and copy what works without getting laughed out of the room. And what's more, I think people would be correct to laugh that person out of the room.

3:35It's not that there isn't going to be a painful trial and error process, but the nature of generative AI in general means that there isn't really going to be a moment when some people have quote unquote adopted it and some people haven't. Generative AI's capabilities are going to increase faster than organizations' abilities to integrate them, meaning that if anyone hangs back and just hopes to wait for best practices, they're likely going to just end up copying something that's out of date by the time they even adopt it. Tanija again from Visa discussed how the firm is managing to move so fast, stating that strong governance infrastructure, data protections, and guardrails are key.

4:05Beyond that, their secret has been to tap the involvement of the teams looking to integrate AI rather than just relying on a smaller group of AI experts, once again reinforcing that this is a total organization-wide bottoms-up effort. He said, this is something that the collective organization has to embrace. Today's episode is brought to you by Vanta. Whether you're starting or scaling your company's security program, demonstrating top-notch security practices and establishing trust is more important than ever. Vanta automates compliance for ISO 27001, SOC 2, GDPR, and leading AI frameworks like ISO 42001 and NIST AI risk management framework, saving you time and money while helping you build customer trust.

4:44Plus, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing trust center all powered by Vanta AI. Over 8 ,000 global companies like LangChain, Lila AI, and Factory AI use Vanta to demonstrate AI trust and prove security in real time. Learn more at vanta.com slash nlw. Learn more at vanta.com slash nlw. That's vanta.com slash nlw. Today's episode is brought to you by Super Intelligent. Every single business workflow and function is being remade and reimagined with artificial intelligence. There is a huge challenge, however, of going from the potential of AI to actually capturing that value.

5:23And that gap is what Superintelligent is dedicated to filling. Superintelligent accelerates AI adoption and engagement to help teams actually use AI to increase productivity and drive business value. An interactive AI use case registry gives your company full visibility into how people are using artificial intelligence right now. Pair that with capabilities building content in the form of tutorials, learning paths, and a use case library. And Superintelligent helps people inside your company show how they're getting value out of AI while providing resources for people to put that inspiration into action.

5:55The next three teams that sign up with 100 or more seats are going to get free embedded consulting. That's a process by which our Superintelligent team sits with your organization, figures out the specific use cases that matter most to you, and helps actually ensure support for adoption of those use cases to drive real value. Go to bsuper.ai to learn more about this AI enablement network. And now, back to the show. Moving over to the regulatory sphere, one of the big trends we've been seeing is, of course, the hyperscalers looking to cozy up with nuclear power as they search for more energy for their AI efforts.

6:29But now, U.S. regulators have blocked Amazon's nuclear power agreement. The Federal Energy Regulatory Commission, FERC, voted two to against a proposal to increase the amount of nuclear power supplied to a data center in Susquehanna, Pennsylvania. This is an already operational reactor owned by Talon Energy, rather than relating to Amazon's small modular nuclear reactor investments. The proposal would have seen the load capacity provided directly to the data center increase from 300 megawatts to 480 megawatts. Utility owners American Electric Power Corp and Exelon Corp had filed a complaint opposing the proposal, stating that it would threaten grid stability and raise customer rates.

7:03The grid operator had warned that they have received over 8.5 gigawatts of large load requests and face an electricity shortfall by 2030. Regulators for their part said that the plan didn't adequately prove why the special contract should be allowed under federal rules and could set a precedent moving forward. FERC chairman Willie Phillips dissented, stating that the grid operator had addressed reliability issues and called the order a step backward for electricity reliability and national security. The decision came after a day-long technical conference on the merits and issues of co-located data centers that draw power directly from generators without using the grid.

7:33Paraphrasing Phillips Bloomberg wrote, Artificial intelligence and related technologies represent a generational opportunity for national security and economic growth. Data centers are driving potentially unprecedented growth in U.S. electricity usage, and the concern is that such deals will allow them to shunt costs to other consumers. However, in a press release, Talent Energy warned, the decision will have a chilling effect on economic development in states such as Pennsylvania, Ohio, and New Jersey. They commented that their proposal is actually part of the solution to issues raised by the commission, adding, it brings service to the customer quickly and without expensive transmission upgrades necessary to serve large load demand.

8:05But our Direct Connect configuration is just one of several commercial solutions to the demand of large loads, and we are exploring other solutions as we move forward. The data center economy will require an all-of-the-above approach to satisfy the increasing demand, including co-locations such as Talon's arrangement with AWS, hybrids that co-locate primary power behind the meter while using grid meter for backup, and front-of-the-meter connections to utility transmission. Staying on the theme of the extreme energy and compute costs surrounding AI. AI cloud startup CoreWeave told investors revenue is set to skyrocket thanks to a big new deal with Microsoft.

8:36They said that the hyperscaler plans to spend$10 billion with them between 2023 and the end of the decade. CoreWeave said they currently have signed contracts worth$17 billion in total across all customers and expect to see revenue of$2 billion this year. They expect revenue to quadruple to$8 billion next year. This news comes days after Microsoft's earning report where they admitted their cloud division is supply constrained and would need to construct new data centers to increase capacity. CoreWeave deals exclusively with AI training and inference, but a boost in the segment could help Microsoft keep up with the deployment of AI applications.

9:05Speaking of Microsoft, Microsoft and venture firm A16Z have set aside their differences to petition the government for more permissive startup policy. In a joint letter, they wrote, Our two companies might not agree on everything, but this is not about our differences. It is about jointly recognizing that the policy choices or missteps we make now will determine whether the U.S. can continue our long and proud history of fostering innovation and seeing startups, small businesses, and entrepreneurs succeed. After all, we both know a thing or two about the little guy working to achieve greatness from their garage.

9:31That's the story of Microsoft and the mission of A16Z. The letter is a continuation of A16Z's work on promoting startup-friendly policy, which they've called the Little Tech Agenda. The firm has already provided staunch opposition to regulations, such as California's SB 1047, which they viewed as giving an advantage to big tech. The letter included a number of policy recommendations, including promotion of open-source innovation and the creation of open data commons for trading purposes. They suggest the government should regulate AI with a view to maximizing competition and choice, stating, Regulators should not only permit providers to offer a broad array of models, proprietary and open source, large and small, but should permit developers and startups the flexibility to choose which AI models to use wherever they are building solutions, and not tilt the playing field to advantage any one platform.

10:11When it comes to the contentious issue of copyright, they wrote, Copyright law is designed to promote the progress of science and useful arts by extending protections to publishers and authors to encourage them to bring new works and knowledge to the public, but not at the expense of the public's right to learn from these works.

10:35Overall, they asked the government to take a quote, science and standards-based approach that recognizes regulatory frameworks that focus on the application and misuse of the technology. Regulation, they said, should be implemented only if its benefits outweigh its costs. In accounting for costs, policymakers should include an assessment of possible costs associated with unnecessary bureaucratic burdens to startups. As the new global competition in AI evolves, laws and regulations that mitigate AI harm should focus on the risk of bad actors misusing AI and aim to avoid creating new barriers to business formation, growth, and innovation.

11:06Lastly today, given the U.S. elections coming up, Perplexity has launched an AI-powered election hub. The service will provide AI-generated answers to voting questions and summaries of candidates ahead of the vote. Once the polls are closed, the hub will serve live vote counts as they come in across the country. The Verge is a little skeptical, writing, perplexity is putting to the test whether it's a good idea to use AI to serve crucial voting information. They managed to find a few flaws, such as a joke write-in candidate called future Madame POTUS, appearing with meme pictures of Kamala Harris.

11:32A spokesperson said the startup is working very hard to get this right, commenting, We selected domains that are nonpartisan and fact-checked, including Ballotpedia and news organizations. We're actively monitoring our systems to ensure that we continue to prioritize these sources when answering election-related queries. I will certainly be checking this out to see how it works, and I will share a report later in the week. For now though, that is going to do it for this extended headlines edition of the AI Daily Brief. Appreciate you listening as always, and until next time, peace.