Microsoft, @satyanadella please do something about this! No option for thinking takes you to the bottom of the list

MS dropping some cool features in copilot lately (agent flows, app builder). They are limited to frontier, but still not available on my account. @Copilot can you help out here? Frequent challenge with your announcements is working out whether I can actually test these

Humanoids were long a thing of sci-fi, then they were a thing of research, but today, with the launch of NEO, humanoids become a product. NEO is the first step on a journey towards a more abundant future and we’re excited for you to join us on this journey. Order your NEO today.

Google Earth AI, our collection of geospatial AI models and datasets, is expanding globally and adding new capabilities. That includes Geospatial Reasoning, powered by Gemini, which automatically connects different Earth AI models - like weather forecasts, population maps + satellite imagery - to answer complex questions.  We’re also bringing new Earth AI models to Gemini capabilities in Google Earth, which make it easy to instantly find objects and discover patterns from satellite imagery. For example, analysts could spot harmful algae blooms that could impact drinking water supply, and issue warnings.

A small audio model launch -- gpt-4o-transcribe-diarize This is a diarization-focused ASR model, it's big and slow so we recommend running it offline, but it excels at differentiating speakers, and you can provide voice samples for known speakers up front.

"Hashmi has declined all debates and will be represented by artificial intelligence."

New paper: LLMs are increasingly used to label data in political science. But how reliable are these annotations, and what are the consequences for scientific findings? What are best practices? Some new findings from a large empirical evaluation. Paper: eddieyang.net/research/llm_a…

Claude can now use Skills. Skills are packaged instructions that teach Claude your way of working.

A lot of problems with AI discourse are because "being good at AI" (called Theory of Mind in this paper) is a skill that seems to be independent of "being great at your job" So you have amazing experts who gain from AI, and others who do not, and they don't understand each other

One AI year is seven Internet ones.

not enough people are emotionally prepared for if it’s not a bubble

AI agents are utterly changing software project timelines right now, and thus what kind of work we can start to work on. The kind of projects that used to be estimated at months are now being scoped to days or weeks. And when they’re not, it’s usually just because we’re not being creative enough or just pushing AI enough. It’s kind of crazy the speed at which the entire practice of engineering is changing. The impact is that our software over time will be able to just be vastly more capable and useful than it is today. Historically, the majority of time in product development gets chewed up on the most urgent, but often least interesting, areas of work. Fixing bugs, upgrading a library, improving performance, improving a feature from a customer request, and so on. You’re always aiming to carve out more time to do the bigger ideas, but it’s very hard. Now, because you’re able to just work through the obvious backlog of tasks and projects far faster, you now get room to deliver on the things that you never would’ve gotten around to otherwise. And because AI makes it far faster to prototype, you can explore ideas that would be hard to prioritize, but now you can see how useful they end up being far faster. This has pushed the ability to experiment by like 10X on projects because the cost of failure is much lower. As a result, it’s much easier to say “yes” to the kind of ideas that always fell by the wayside before. Interestingly, this lowering of the barrier to building more will mean that there’s an even higher premium on good product management and design. In a world of infinite choices, determining what to build - and how it should look and work - is going to matter more and more.

Friday evening reflection: Adjusting my cloud infra priors for the AI era… a colleague just shared the progress we are making with our AI WAN. In one project, we expanded our North American optical fiber footprint by 40% and added network capacity equal to one-fifth of our entire global network - a network that took 15 years to build now takes months!

Today we launched Tinker. Tinker brings frontier tools to researchers, offering clean abstractions for writing experiments and training pipelines while handling distributed training complexity. It enables novel research, custom models, and solid baselines. Excited to see what people build.

My test of any new AI video model is whether it make an otter using wifi on an airplane Here is Sora 2 doing a nature documentary... 80s music video... a thriller... 50s low budget SciFi film... a safety video.. film noir... anime... 90s video game cutscene... French arthouse

if i had to choose a single 5-minute read to agi-pill smart friends, it would be this clean and simple piece

AI coding agents offer the best glimpse into what the future of agents will look like in many other fields of knowledge work. AI coding has accelerated faster than any other AI space because the builders of AI understand their own workflows deeply, and they’re incentivized to improve it for their own productivity. It’s also a great petri dish because the ecosystem will adopt new tools (and dump old ones) faster than other space, which gives you better signal faster to what paradigms work and which don’t. The darwinian forces are very strong here. While there are plenty of things that don’t translate from coding to other areas of knowledge work, in agents it’s clear that we’re starting to see the formulation of the core primitives for agents in knowledge work. These foundations include: background agents that you kick off in a simple GUI or via another logical trigger, the ability to track their progress, add in relevant context, ability to pull in additional signal and tools, ways of reviewing the work and output at the end of the workflow, creating custom agents on the fly when workflows are repeated, and so on. We’re in the earliest phases of what this will look like across software, but the fuzziness that we had a year ago is starting to get a bit more clear by the day. Incredible times ahead.

Critical read, and it doesn't just apply to startups. So many problems in more established legacy businesses could be solved by following this advice.

How many hundreds of millions wasted because orgs thought they could do better product or couldn’t get AI past compliance?

Been telling clients this for 2 years now. The competition is coming, it’s going to be lean and incredibly adaptable