HW: Tensormesh was founded by AI systems researchers from the University of Chicago, UC Berkeley, and Carnegie Mellon, led by Professor Junchen Jiang, co-creator of LMCache, one of the leading open-source KV caching projects. The company's core insight is simple: as AI applications move into production, inference, not training, becomes the biggest cost driver. Most AI systems repeatedly process the same context, prompts, and workflows, wasting GPU resources every time. Tensormesh solves this problem through KV cache infrastructure that allows AI systems to reuse previously computed results instead of recomputing them from scratch, reducing latency and GPU costs by up to 10x. If Together AI is building the cloud for open-source AI, Tensormesh is building the memory layer for AI inference. Tensormesh is betting that the future bottleneck of AI won't be intelligence, but the cost of repeatedly running intelligence.

Yep. I keep coming back to this: you can’t skip the basics and jump straight to sovereign AI branding. If universities aren’t doing the unglamorous post-training work, it’s mostly vibes.

Unclear to me if the wolf domestication analogy really works, but the selective pressure point does. Buyers want capable models, not cyber incident machines.

Hermes Agent updates (the most important ones) • Hermes turned from a “CLI agent” into a full desktop app. • Hermes overtook OpenClaw on OpenRouter • It became faster and more modulAR • Anthropic / HERMES.md billing incident • Some developers moved to Hermes because of OpenClaw vulnerabilities We’ve updated our Hermes article with the latest news – read the whole story is in one place: turingpost.com/p/hermes

Extremely exciting regardless, but "one tool, ten minutes, $10k/month" is very caffeine-coded ☕️

So I have trouble agreeing with the takes that these budget cuts are bad new for the AI labs The reason the budget caps are necessary is that coding agents got undeniably useful - that's a complete turnaround from a year ago

🚨breaking: bad news for Anthropic since Meta was said to be a big customer and is cutting its token budgets. more generally lots of companies will make the same decision; next year’s token budgets won’t be the freewheeling affair they were earlier this spring. honeymoon is over. in a rational world a lot of companies would take a hit on this news.

Timestamps: 00:20 — Intro: Ivy’s role at Google Research & Gemini for Science 01:00 — What Gemini for Science actually launched at Google I/O 02:45 — Current state of AI for Science 04:45 — Why scientific agents are fundamentally harder than coding agents 07:25 — Hallucinations, data leakage, and how AI agents can “cheat” 08:40 — Challenges of training AI for specific scientific verticals 09:40 — “Catch it early”: using LLMs as judges to detect when scientific agents cheat 12:15 — Ivy’s hot take: most AI demos are cherry-picked

What does it really take to build AI that can do science? Fresh off the Gemini for Science launch at Google I/O, Google engineer Ivy joins us to talk agents for science: the new tools coming for researchers, and why science is so much harder for AI than code. Timestamp in the thread 🧵

The interesting part to me isn’t even the “money glitch” angle, it’s whether this makes some workflows actually productizable. If yes, that’s a pretty big deal.

Diane (Anthropic, shipped 21 versions of Claude): "Last night, while we were all asleep, a product noticed it was broken. It read its own error reports, found the bug, wrote the fix, and rolled it out. By the time the team woke up, the problem was gone." That is what a self-correcting agent looks like in production. Fable 5 verification loops outcomes rubric. 9 out of 10 agents today fail silently and report "done." The team that builds Claude is already running agents that catch and fix their own mistakes overnight. Read the 5-step build below. Worth more than a month of manual checkpoints.

Google DeepMind CEO Demis Hassabis on why young people should make themselves "superpowered" with AI tools: When Hassabis gives talks at universities and schools, his core piece of advice is simple: Lean all the way into where technology is heading. "You've got to just go with the flow of the direction," he says. "I would immerse myself in every tool available and just become almost like superpowered with those tools and those capabilities." His reasoning comes from what he sees inside the frontier labs themselves. So much effort goes into simply building the next versions of these models that even the people making them can't keep up with everything those models could do. "Even at the frontier labs, there's so much work that has to go into just making the next versions of these frontier models and then all the adjacent models. So for us, like Veo and Nano Banana and Gemini, even we can only explore a fraction of the applied things you could do with it, the applications you could make with it." And that gap, @demishassabis explains, is widening: "I think that gap's getting bigger and bigger, in terms of the overhang of the capabilities, all the cool stuff on the latest models. And the release schedules are getting faster and faster." This is where Hassabis sees the opportunity. The people positioned to win aren't necessarily the ones building the models, they're the ones who master the tools and point them at something new. "The opportunity space is getting huge for people who are really expert at using those tools and then apply it to some new domain." He puts it in stark terms: "A kid these days could probably start a multi-billion dollar business in some ways, using these tools in some new way that no one had thought about."

Dear Canva, I have bad news 😭 This agent can now take one random product idea and turn it into an entire brand kit. Packaging. Ads. Carousels. Pitch deck. Everything. I tested CapCut Design Studio and this is not normal:

We are broadening the scope of the blog posts on MathArena, might want to extend it further to allow for all kinds of analyses on AI4Math, not sure how exactly it should look like though.

Totally fair red line. 'Please upload your crown jewels, we may keep them' is a pretty wild product pitch.

How does beam search work? An LLM generates text by computing the probability of a word coming up, for example given the text "what a nice" the LLM might predict "dog" with probability 50%, "cat" with 40%, and every other word with <1% probability. 1/

I made a personal black hole that makes you take breaks 🕳️ A shader for Ghostty that spawns a small black hole in your terminal - it drifts around, gravitationally lensing your text. The longer you work without stopping, the bigger it gets, until it's basically demanding you go touch grass Take a break and it quietly shrinks away

my favorite genre of open model is 'fully local' followed by a casual 19.5GB quant

The Abu Dhabi Investment Forum Shanghai 2026 was held in the Chinese city today, focusing on cutting-edge fields, including AI, e-sports, advanced manufacturing, cross-border investment, and experiential economy. The event provides a high-end platform for business leaders, investors, and family offices to promote in-depth exchanges and explore new global opportunities. #ADIF2026 @ADGlobalMarket @InvestAbuDhabi

NIP Group chose Abu Dhabi as its global HQ because of the city’s position as the crossroads of East and West, and the UAE’s national vision for sectors, such as digital assets and AI, Hicham Chahine, founder and co-CEO of the Chinese entertainment firm, said at #ADIF2026 today. #This_is_AbuDhabi #InvestwithAbuDhabi @InvestAbuDhabi