Tom

Ok this is cool

@chrisfralic @elevenlabs @Carles_Reina

Who do I know that’s an investor or works at @elevenlabs - I’ve got a cool important project that could use their tech.

A strong signal of enterprises that can move the fastest with AI - they have in-house data engineers, scientists, and analysts that can lead AI programs. If a company outsourced data to consultants, they’re typically ~6-12 months behind as they need to hire before they can build

AI can now generate scientific ideas at scale. But we need to know if the current state of the art can bridge the gap to physical validation – the phase constrained by what can be tested, how fast, and at what cost. To find out, we have doubled our investment in the AI Scientist programme to £6m. We're backing 12 projects to see if autonomous systems can reason, plan, and run experiments in the real world. These teams are testing the limits of automation on deliberately unforgiving problems: Alzheimer’s and cancer therapeutics, material discovery, and understanding the mechanisms behind battery degradation. Instead of looking for best-case scenarios, we’re looking for limits. Can these systems recover when experiments fail? Can they reason across disciplines? Can they decide what not to try? By doing this, we are learning what happens when machines are asked to do science, and exploring what that means for the future of discovery. Discover the projects: link.aria.org.uk/AIscifpx

I've always thought rules killed speed and that this explained the UK’s lack of growth. When decision latency will approach near zero, governance and velocity will stop being trade-offs. Rules and regulations will become guardrails and inference will unlock them

A year ago, we verified a preview of an unreleased version of @OpenAI o3 (High) that scored 88% on ARC-AGI-1 at est. $4.5k/task Today, we’ve verified a new GPT-5.2 Pro (X-High) SOTA score of 90.5% at $11.64/task This represents a ~390X efficiency improvement in one year

@GergelyOrosz Slack had a video calling feature called Slack Calls that got deprecated in 2022 in favour of audio only Huddles. At the time Zoom was eating all competition.

How is it that despite Slack being (probably?) the most loved chat tool, it has no usable video calling product? All companies I know using Slack use Meet / Zoom / or even MS Teams for video calling. Why did Slack never build anything? (no, Slack Huddles does not count as one)

@BarrAlexandra Either Sydney at one of best new customers or in your MB2 lair!

@TomPrickett Where is the second pic? 👀

@matthewclifford Congratulations and thanks @matthewclifford 🫡

Today was my last day at No 10 as the Prime Minister’s Adviser on AI. It’s been a privilege to serve over the last year and - I hope - to make a small contribution to putting the UK on the path to being an AI winner

@thegautam So cool

TL;DR: We built a transformer-based payments foundation model. It works. For years, Stripe has been using machine learning models trained on discrete features (BIN, zip, payment method, etc.) to improve our products for users. And these feature-by-feature efforts have worked well: +15% conversion, -30% fraud. But these models have limitations. We have to select (and therefore constrain) the features considered by the model. And each model requires task-specific training: for authorization, for fraud, for disputes, and so on. Given the learning power of generalized transformer architectures, we wondered whether an LLM-style approach could work here. It wasn’t obvious that it would—payments is like language in some ways (structural patterns similar to syntax and semantics, temporally sequential) and extremely unlike language in others (fewer distinct ‘tokens’, contextual sparsity, fewer organizing principles akin to grammatical rules). So we built a payments foundation model—a self-supervised network that learns dense, general-purpose vectors for every transaction, much like a language model embeds words. Trained on tens of billions of transactions, it distills each charge’s key signals into a single, versatile embedding. You can think of the result as a vast distribution of payments in a high-dimensional vector space. The location of each embedding captures rich data, including how different elements relate to each other. Payments that share similarities naturally cluster together: transactions from the same card issuer are positioned closer together, those from the same bank even closer, and those sharing the same email address are nearly identical. These rich embeddings make it significantly easier to spot nuanced, adversarial patterns of transactions; and to build more accurate classifiers based on both the features of an individual payment and its relationship to other payments in the sequence. Take card-testing. Over the past couple of years traditional ML approaches (engineering new features, labeling emerging attack patterns, rapidly retraining our models) have reduced card testing for users on Stripe by 80%. But the most sophisticated card testers hide novel attack patterns in the volumes of the largest companies, so they’re hard to spot with these methods. We built a classifier that ingests sequences of embeddings from the foundation model, and predicts if the traffic slice is under an attack. It leverages transformer architecture to detect subtle patterns across transaction sequences. And it does this all in real time so we can block attacks before they hit businesses. This approach improved our detection rate for card-testing attacks on large users from 59% to 97% overnight. This has an instant impact for our large users. But the real power of the foundation model is that these same embeddings can be applied across other tasks, like disputes or authorizations. Perhaps even more fundamentally, it suggests that payments have semantic meaning. Just like words in a sentence, transactions possess complex sequential dependencies and latent feature interactions that simply can’t be captured by manual feature engineering. Turns out attention was all payments needed!

Agent Tools for Developers openai.com/live/

Sora has arrived in the EU and the UK.

Breaking: OpenAI just released GPT-4.5, the startup's largest AI model to date. Available now to Pro ($200/mo tier) users and developers on paid tiers via API. Everything else you need to know about the highly-anticipated launch:

@athyuttamre COOKING

The OpenAI API team is cooking. Can't wait to for our next set of products to launch.

1/ Today the Prime Minister published my AI Opportunities Action Plan for the UK and committed to implement all its recommendations. A quick thread on why this is important, what we need to do now - and how you can help

We’re looking for people to test o3 and o3-mini, want to help? openai.com/index/early-ac…

We announced @OpenAI o1 just 3 months ago. Today, we announced o3. We have every reason to believe this trajectory will continue.

Day 10 🎁: 1-800-CHATGPT ✨ Call ChatGPT on your phone if you're in the US, or message it on @WhatsApp from anywhere on the planet. It's free, no account needed! Give it a try right now.

The OpenAI outage was unrelated to 12 Days of OpenAI or Apple Intelligence. We made a config change that caused many servers to become unavailable (the main impact window was 3:16–7:38pm PT). We’ll share a postmortem at status.openai.com/incidents/ctrs… after the investigation is complete.