Your curated collection of saved posts and media
The first day in San Francisco has been incredible. New friends. My rock is on the way. I bought some flowers. Got some more tiny plates for our new friends. https://t.co/0z20RyTP0w
I’ve been living immersed in South Korea for the last month with my wife and daughter. We just leave the stroller outside if we go in anywhere. Because no one’s going to steal it. I met a friend for lunch. He rode his bicycle. He just left it outside in an alley without a lock. It was still there 2 hours later. Another friend mistakenly left his phone on a park bench. When he finally retraced his steps and went back 4 hours later, his iPhone was still there. I walked by a KPop concert. The fans who traveled from outside of Seoul to attend just left their luggage outside the subway station. No locks. No security. Koreans take this for granted. They don’t realize this is not normal for most the world, especially America. When I ask about it, they just respond, “of course, why would someone take it?” Can you imagine any of these things happening or being possible in NYC or LA or *insert city*? And IF, something were to be stolen in Korea, the police would investigate. Because in a high trust society, rules and norms matter. There’s no “under $1000” law. Theft is theft. And trust is trust. Will this social norm ever be possible in America?
https://t.co/pBWhE53c4A
https://t.co/pBWhE53c4A
"When an LLM outputs a step-by-step plan, it creates a powerful illusion that you are watching a machine reason its way to a conclusion. A position paper by professor Subbarao Kambhampati and researchers at Arizona State University systematically dismantles this assumption." (From @bendee983 via @bdtechtalks ) 👉https://t.co/ELPfUplpU0
Falcon 9 launches 24 @Starlink satellites from California https://t.co/UmMBikZgK5
Brazilian friends, did I do good? A highlight of Rio was spending an hour in the record store listening to tracks on each album and giving the owner a thumbs up or thumbs down. I’ve had Di Melo, Joutro Mundo, Rosinha de Valença for years, a gift from my neighbor in Bondi Beach. https://t.co/ceNvAGMZGF
At @CERN, finding the universe's smallest particles requires big collaborative efforts. 🤝 Research fellow Batoul Diab shares how the ALICE collaboration uses open source code on GitHub to analyze massive amounts of physics data. Watch to see shared code, peer review, and global teamwork power modern scientific breakthroughs. 💡
🐕🦺 Blessed beyond measure to have gotten to spend yesterday afternoon and evening brainstorming Pupper for Good deployments with these brilliant doctors and researchers!! Teresa Nguyen is such a polymath and a powerhouse, as are Claudia Mueller, Laura, and Stuart. @googleaistudio @googledeepmind @googlegemma ftw!💎
We actually lived through this https://t.co/daL3jUqS6i
OpenClaw v2026.6.9 is out, with a focus on paper cuts! 💬 Richer Telegram delivery 👏 Steadier agent recovery 🧬 Stronger Codex integration 📦 Slimmer distribution 👌 Improvements in search and skills https://t.co/GOoHDXU8MZ
GLM 5.2 is now on DeepSWE as the top open-source model on our leaderboard. With a pass@1 score of 44% at max effort, GLM 5.2 is indisputable #1 open-source model besting Kimi K2.7 Code by 17%. https://t.co/cYZBm5z909
Pretty dope @greptile drink at @ycombinator https://t.co/BQ9BtrIcfu
>> Scalable Evaluation for AI Agents << If you run agent evaluation in production, this one is worth your time. It shows that front-loading human judgment into reusable evaluation assets is useful. But why? Agents reason across turns, call tools, hold context, follow policies, and act under uncertainty, so they have to be judged as behavioral systems. Current methods each give a fragment. Benchmarks measure fixed capabilities, human review preserves judgment but does not scale, LLM-as-judge inherits the evaluator design problem, red teaming is episodic, and trace audits need explicit evidence rules. Human-on-the-Bridge puts human expertise upstream, where experts curate reusable evaluation intelligence before testing rather than reviewing each output in the loop. Paper: https://t.co/0dVOH3QrZ6 Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX
I love this guy. Muslim former Prime Minster of Scotland. When a white person attacks brown people, he goes into a frenzy posting about racism and hate. When a brown person attacks a white person, he calls for an end to "division". Same act. Two standards. Anti-white racism. https://t.co/hHWcvfZVw4
My full statement on the horrific, but I am afraid to say, unsurprising attacks in Edinburgh. Enough of the statements of solidarity from governments. Muslims - across the UK - no longer feel safe in the only country they call home. Time to face down the peddlers of hate. http
For Father’s Day, I’ll share this letter to Karl Marx from his dad https://t.co/m8WHi4LhRz
Traded my Range Rover for a @Tesla Model Y and picked it up two weeks ago. Turned on FSD today after the 14.3.3 update. If you don’t believe in magic, I don’t know what to tell you. I’ve always been fascinated by technology for as long as I can remember, but this is something else entirely.
No Taxation Without Annihilation The Anti-Antimatter-Tax Party fights for our freedom. AATP LFG! https://t.co/fJMpnMmb3x
Happy Father’s Day to all the incredible dads out there ❤️ https://t.co/ZZURvmfEWj
Happy Father’s Day to my wonderful sons, my fantastic father and all fathers. 🥰🥰 https://t.co/pkpNiYStwv
As India builds frontier AI rooted in its own languages and knowledge, IIT Bombay and BharatGen are proud to support and participate in Project Tapestry: an open, global consortium for nations to advance frontier AI together. BharatGen, supported by the IndiaAI Mission and the Department of Science and Technology, joins as a founding contributor with multilingual AI infrastructure built natively for India's diverse languages. A Letter of Intent was signed on June 18th by IIT Bombay Director Prof. Shireesh Kedare, in the presence of both Deputy Directors Prof. Milind Atrey and Prof. Ravindra Gudi, and Prof. Ganesh Ramakrishnan, Founding Director at BharatGen. Project Tapestry was represented by Dr. Christopher Nguyễn, Chief Architect of Project Tapestry and AI Alliance Board Member. What India has built for its own people, it now brings to the world. @EduMinOfIndia @BharatGen_Com @ganramkr @OfficialINDIAai @IndiaDST @ylecun @kb_bha @pentagoniac @kb_bhatta
Same page, two captures. https://t.co/BQWx8PIy2M as PNG: 431 KB https://t.co/BQWx8PIy2M as WebP: 74 KB One flag: --format webp. 83% smaller. If your agent stores thousands of grabs, this is the cheapest optimization you will ship today. https://t.co/UoBINkAZ9i
Same page, two captures. https://t.co/BQWx8PIy2M as PNG: 431 KB https://t.co/BQWx8PIy2M as WebP: 74 KB One flag: --format webp. 83% smaller. If your agent stores thousands of grabs, this is the cheapest optimization you will ship today. https://t.co/UoBINkAZ9i
What literatures have developed since the paper that are in dialog or tension with its themes or findings? (This is something that is very hard for academics, ensconced in a field, to do on their own) https://t.co/TX3gdVqnbP
🚨 New DeepMind’s Category Error on Transformer “State” (Mozer et al. 2026) Paper correctly flags transformers’ sequential consistency limits and offers a useful recurrent taxonomy. Core claim: “depth exhaustion” from pushing state deeper, misreads the transformer architecture. The residual stream isn’t a persistent state structure or evolving “mental model”. It’s an intermediate calculation within the single pass forward based on token isomorphism and training corpus regularities devoid of any system-wide coherent meaning but context dependent and fragile per inference: h0 = prefill(prompt) - - - - - - - - - h1 = h0 + layer0 h2 = h1 + layer1 . . . / layer_i = Attn(i) + FFN(i) / . . . - - - - - - - - - A running sum across layers (not time). The full stack computes hL once, then last_FFN + unembed → logits → one token. - - - - - - - - - t = unembed(last_FFN) - - - - - - - - - That token alone seeds the next auto-regressive step. - - - - - - - - - input = input + t - - - - - - - - - No internal carry-over. No “beliefs.” Just partial aggregates overwritten in the next pass. “Depth exhaustion” is an artifact: treating mid-layer partial sums as final outputs. Reading tea leaves without tasting. Intermediates at layer 6 aren’t disambiguated interpretations, they’re provisional values further refined by remaining layers. Many paths can converge to the same final token. Only the emitted token is causally decisive. Chain-of-thought isn’t a hack. It’s the only state mechanism: tokens are the architecture’s persistent memory. Proposing hidden recurrent dynamics just buries state in unauditable vectors while ignoring the real bottleneck, training data for trajectory coherence. The math hasn’t changed since Vaswani 2017. The field has: rewarding narratives of “interpretable belief states” over the simpler truth of ephemeral partial sums in a fixed approximator. Incentives > clarity Non-anthropomorphic Terminology refresh for precision: • “State tracking” → trajectory coherence across autoregressive steps not single pass layers • “Belief at layer N” → intermediate residual stream calculation sum reachable by many paths •“Depth exhaustion” → attention softmax competition under pretraining density •“Implicit reasoning” → unobservable vector transforms Mechanistic Interpretation unfalsified anthropomorphic assumptions are leading the whole field astray.
Happy Father’s Day to all the dads and grandfathers out there helping inspire and prepare the explorers, builders, and dreamers of tomorrow. The future Starfleet is going to need a lot of recruits! https://t.co/jyM4SgkofC
