Yosyp Schwab

@jxnlco rental car?

We going to New York! Found 4 strangers to split with.

@thsottiaux ability to turn on remote control for sessions between codex cli, desktop, and ios app

How can we make ChatGPT better. It’s already on your phone, but what do you wish would happen more or less of when you open the app. How can it bring you more joy?

1) can close the laptop lid and the vibing continues 2) codex cli uses the local filesystem to store work in progress, not /tmp, which i can git branch and have multiple independent sessions working on the same codebase 3) store shell environment variables, big unlock as it enables one-shot "deploy this work to my aws account as dev (don't load balance, save cost, use static S3 site or add containers to the existing ECS and EKS) and host it on this subdomain"

It’s surprising how many people I’ve talked to still use the terminal over the app. why

@jun_song instructions unclear. should i start it with `ollama run kimi-k3` or `vllm serve moonshotai/kimi-k3`?

Guide on how to run Kimi-K3 locally : - you need to buy 10 x GB300 (252GB VRAM each), thats about $1M - Electricity bill estimated $10k/month

@lilyjclifford the full benchmark dataset is on huggingface. the benchmark score is crowdsourced by @arena users voting for their preferred model output on the same prompt. huggingface.co/datasets/lmare…

benchmarks are so broken. if you search 'frontend code arena benchmark' - you just get news about kimi k3, no details on what the actual benchmark is. that's the definition of broken

@SemiAnalysis_ Big if true

BREAKING: Broadcom is diversifying away from long-time foundry partner TSMC. Lego will be the new manufacturing partner for Broadcom's flagship hyperscaler ASIC products in 2028. Broadcom President of Semiconductor Solutions, Charlie Kawwas, has already showed off package samples at RAISE Summit in Paris last week. With TSMC supply constrained, customers are alternative sources of chip supply from unexpected places. One of the major differentiators Lego brings is industry leading defect repair, "plug and play" chiplet interoperability, as well as built in self-alignment technology for 3D stacking. Mattel and Hasbro were also under consideration, but Lego's proven track record won out in the end. We have ordered multiple Millennium Falcon Sets set to our Portland teardown lab for competitive teardown analysis.

@chamath almost. not all tokens are created equal. @ArtificialAnlys tracks this by "output tokens per task". comparing that benchmark vs cost $ gives the clearest picture imo

The math ain’t mathing. What this shows is that you can now spend $0.50 per 1MM leading edge tokens or spend $56 for those same 1MM tokens. 🤔

@Saas_addy @HeddenMorgan

I'll be back in 10 minutes’ is the biggest lie every developer has ever told. 💀

@depletionmode @J_Alfred_Pruf @InsiderPhD see also: dl.acm.org/doi/10.1145/35…

Sure. To an extent. But… in the context here - how do we define verifiable trust? Also increased costs, latency etc. I’d imagine that tactical mitigation eventually may look like some sort of classifier or something on top of each inference cycle to check tool calls, weigh vs intent etc. But a lot more research is needed in the field to properly map and evaluate.

🧵Can we trust Chinese open weight models? Was a question a lot of people asked after GLM 5.2 was released, scoring very well on coding benchmarks, and suspiciously Claude-like. So I turned an open-weight coding model into a backdoor with 1hr and <$100. Let's talk about it

@foresightinst Congrats @jesseposner!

@predict_addict Did you use LaTeX typesetting? A preview and table of contents in Amazon would be great!

After months of work, it's live. Andrei Andreyevich Markov finished Calculus of Probabilities in 1900. For 125 years there was no complete faithful English translation. There is now — and I'm proud to have made it. Why this book matters beyond the history: the stochastic processes Markov formalised here — Markov chains — sit beneath Google's PageRank, hidden Markov models, reinforcement learning, and every autoregressive language model running today. In 1913 he turned the same methods on Pushkin's Eugene Onegin, counting 20,000 letters by hand to model the transitions between vowels and consonants — the first statistical model of language ever recorded, and a direct ancestor of modern AI. This is the Faithful Edition: Markov's original structure, section numbering, and notation preserved, with the mathematics reset in clean modern typography. Eight chapters, every result proved from first principles. For historians of science, probabilists, students of stochastic processes, and collectors of the classics of Russian mathematics. Now on Amazon 👇 a.co/d/074xT2jv #MachineLearning #ProbabilityTheory #MarkovChains #HistoryOfScience #DataScience #AI

@levelsio ACmaxxing is underrated, it’s next big thing after hydration and CO2 monitoring

Every. Single. Time. They have AC but you can't have AC!

@InsiderPhD great work! what would a benchmark that catches this look like?

@OwenBrakes Source?

Rip every radio out of your phone. It will still leak your inputs. Capacitive touchscreens scan sequentially. Your finger touching the screen affects the local impedance and modulates the emissions. Able to extract keystrokes and pin inputs from 15cm away with ~90% accuracy.

@dylan522p “centre”

I would like to see them TRY STOP THIS

@cdngdev neat! how many tokens did this consume, and how long did the agents take to complete?

i gave 5.6 sol access to my camera roll and had it extract pictures of every piece of clothing i own from my photos then, told it to find new outfits for me and render them on me with gpt-image! its kinda cool to see your entire wardrobe in a collection like this

@noahzender This is exactly what spending 40 hours on PCPartPicker as a kid felt like. You couldn't afford the parts yet, so you just built the perfect system in your head 100 times over. That obsession is the education.

My favorite story about true mastery: Steve Wozniak designed computers for fun in high school because he couldn't afford to build them. "All I could do was design them on paper and try to get better and better." Contraints helped him develop better skills.

It's mind-blowing how fast agentic coding has progressed in the past 6 month. It's a completely different world now.

@pmarca Measuring doctors on the quality of their charting in Epic is like measuring developer productivity by their Jira tickets and LOC. We need to measure outcomes for AI to gain trust, and patient compliance to prescribed orders is the biggest piece of that.

AI is already a better doctor than 99.99% of human doctors. This is good news.

@antirez automating devops is the highest ROI for programmers to demo what they’re work on with the world. “Use my $AWS_PROFILE generate terraform to deploy this low budget dev project to $SUBDOMAIN and password protect it with simple auth”

It is my belief that many devs right now are not maximizing what they can do with automatic programming because they still look at the code. Doing it makes you the bottleneck. Your time is better invested in new ideas, QA, design, and asking yourself what is your goal.