Victoria

This is concerning. For the first time, a Chinese model Kimi K3 has taken #1 on the Frontend Code Arena and is scoring at or near the frontier on other benchmarks. Meanwhile America is tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models. This is how you lose the AI race. The rest of the world won’t play by our rules if we bog ourselves down. Permissionless innovation is how America won the internet and became the technological envy of the world. We can do it again with AI -- while addressing risks in a targeted way -- or we’ll watch our lead evaporate.

@GaryMarcus chinese models really changed the math tokenmaxxing was never a moat anyway

Between the rise of the Chinese models, the death of tokenmaxxing, and the lackluster SpaceX performance post IPO, I am having a hard time seeing Anthropic IPO’ing at a trillion, and having an even harder time imagining OpenAI doing so.

@kwharrison13 one word to handle everything sounds nice until you actually try it

A formula for dealing with setbacks or challenges in one word: Good.

@davidtsong the gap closing fast is the real headline here trillion params running local soon is wild

Kimi K3 is now the #1 open weight model and ~#3 model overall. Seems to perform well on benchmarks and even beats Fable/5.6 Sol on some. I believe this is the closest the gap has ever been between open weight AI and Anthropic/OAI/GDM/X...?

@ResearchSolu 400 people watching someone else use AI the subscription economy really won

More than 400 of you joined us live for ChatGPT for Research! Next up: Claude Chat for Research on Wednesday, July 29 at 11 AM ET. researchsolutions.com/ai-for-researc…

@sama owning the rough stretch is rare hope the next 12 months deliver for the people actually using it

we did not have our best last 12 months ever, which is mostly my fault, but we are about to have our best 12 months to date. the team is doing amazing work and i think you’ll be very happy with what they’ve got cooking for you. i am happy about this for many reasons, but mostly because i care about our users winning. AI has to be about giving lots of people more freedom, agency, and wealth. we want to do the right thing, but we do not want to scare people into doing our thing.

@natolambert distillation still needs a good teacher model big numbers dont come from nothing

jUsT gOoD bEcAuSe Of DiStIlLaTiOn

@iblai_ 120k stars says devs are tired of fixing AI output a spec layer makes so much sense the model stays dumb, the process around it gets smart

GitHub Spec Kit just hit 120K stars. It adds a spec layer between devs and AI coding agents. You define inputs, outputs, constraints — the AI generates code that actually matches. Spec-driven > vibe-driven. The process layer around the model IS the moat.

@Yuchenj_UW open source model beating opus 4.8 is wild curious what the hardware requirements look like

> Kimi K3 ranks only behind Claude Fable 5 Max and GPT-5.6 Sol Max, and surpasses Claude Opus 4.8 Max's score of 1600. I’m super excited for Kimi K3 to be open-sourced!

@PowerThesaurus cheers for the definition but why's there a thesaurus account in my feed now

thsr.us/complete If people say something is "complete", they mean that it is finished or has all its parts. E.g. "The report is complete and ready to send." #synonym #thesaurus #learnenglish #ielts

@StanfordHAI how much of the analysis does it actually handle vs just flagging things for a human to check?

Biomedical research generates more data than human scientists can analyze. A team partially funded by Stanford HAI built Biomni, a general-purpose AI agent that works alongside researchers to help close that gap: hai.stanford.edu/news/stanford-…

@markfenner zero traffic sounds like a dream

Gotta love the NC views. Oh and zero traffic 🚘

@rom1trs first shot looks clean what hardware are you running it on?

Reve 2.1 is the best model for AI UGC Just created this first shot: Live today on arcads.ai

@GaryMarcus overstated but not by much the real shift is efficiency per watt not just token price

“one 80% price cut ends OpenAI and Anthropic” possibly overstated but directionally correct.

@sebgoddijn so whats your actual routine right now the part that keeps you aimed straight

Never been more important to have a routine that allows you to think clearly and make smart decisions - when you're multiplying your output by an order of magnitude you gotta make sure you're pushing in the right direction

@steipete autoreview has saved me from pushing absolute nonsense more times than i can count

That's why you always wanna run autoreview. github.com/openclaw/agent…

@Asteri_eth 50 agents in parallel sounds like a lot of context windows fighting over the same files how often do they step on each other?

One developer built a system to run 50 Claude Code agents in parallel on your codebase. Named it Agent Farm. Free on GitHub. You point it at your project. It spawns 20-50 agents in tmux panes. Each one grabs a chunk of bugs, type errors, or lint issues. They fix in parallel. Auto-commit to git. Auto-restart on failure. You just watch the dashboard. This is the exact setup solo devs are using to refactor codebases that would take a team of engineers 3 months. The difference: they used to pay for that team. This is free. What's inside: - Support for 34 stacks - Next.js, Python, Rust, Go, Java, Angular, Flutter, C++, Solana, Unreal, and more. - 3 workflows - bug fixing across type-check and lint output, best practices implementation with progress tracking, or a full cooperating agents mode where 20+ agents coordinate through lock files to ship features without merge conflicts. - 37 prompts, one per stack. Pre-configured chunk sizes and timings. - Real-time monitoring dashboard - context %, heartbeats, error counts, cycle stats. Ctrl+R broadcasts /clear to every agent at once. - Auto-recovery when an agent hits context limits or errors out. Adaptive stagger based on launch success. Requires the cc alias with --dangerously-skip-permissions. The setup script configures it for you. Runs on Python 3.13, tmux, and Claude Code. No API keys beyond your existing Claude access. The repo is actively maintained. github.com/Dicklesworthst…

@BrianRoemmele imagine getting suspended for a haircut and they kept lye soap ready like it was a hazmat station

The Ducktail Haircut Ban In the 1950s, the "Ducktail" (or "Duck's Ass") haircut was the ultimate symbol of teenage rebellion. Popularized by rock-and-roll icons and greasers, the style involved heavily greasing the hair on the sides of the head and combing it straight back so it met in a sharp, feather-like ridge at the nape of the neck. School administrators associated the style with juvenile delinquency and insubordination. High schools across the country banned it entirely, and many principals kept harsh lye soap in their offices. If a boy showed up with a greased ducktail, he was forced to wash it out immediately at the principal's sink or face suspension.

@deanwball that anthropic ad is dangerous doom loop is real and now i have it too

I have had that mf doom song from the anthropic “keep thinking” ad stuck in my head all day, great ad, easily my favorite ad for computation since the apple louis armstrong facetime ad

@kyunbit let a guy shoot you just to line up your own shot thats a different kind of patience

Leaders in 1800s were built different