Meet Wraith 👻 the fastest terminal on the web.
👻 Ghostty's terminal engine
⚡ Rust WebGPU renderer, off the main thread
🖼️ Kitty graphics: real images in the terminal
🔤 Ligatures + Arabic & Hindi shaping
🔌 Drop-in xterm.js API, one import
🎮 DOOM in your browser (in the terminal): 400+ FPS
I cannot explain it but here's how I think current models behave best:
opus 5.5: you need to be 'bro' with it. get him on your team and explain the exact intent you're building with, and you'll see some amazing results when you give it ambitious things.
fable 5.2: you need to put it on a pedestal. state that you believe that it knows better than you, and architect without your assumptions, and you'll be surprised by new things it suggests.
GPT Astra/ Sol 6.2: you need to just yell at it and tell it EXACTLY what needs to be hillclimbed. if you don't know how to, ask fable or opus to do it.
you'll be surprised by how novel approaches it can take sometimes to possibly some dead ends declared by claude models.
ymmv
one day you’ll look back at this period and realise you could’ve built anything you imagined.
make sure your future self knows you went for the most ambitious thing possible.
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included).
Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs.
Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines.
This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!!
Enjoy: science.caseyhandmer.com/phy…
I tried some hillclimbing with Sol 6.2 and it seems to be doing a much better job overall.
Obviously these are still not great models to talk to, but they can find novel solutions to problems.
Some results even surprised Fable as it was orchestrating the agents.
One of the funny reasons for the difference in vibes between the influencers and the crowd today on 6.2 Sol/Astra is that we gave a few folks access a while ago and then made a lot of changes in the last 3-4 days and I'm pretty sure no re-testing was done within those days and the videos were produced last week.
Don't get me wrong, while the models are a hefty jump compared to their GPT predecessors, they are by no means perfect and there are a few things the teams are excited for the models to get better at in future iterations in terms of taste and niceness of use. Love all the feedback and we hear you!
GPT-6.2 Sol—an efficient and versatile model for everyday work.
It powers through multistep agentic workflows, writes clearly and naturally, and is intuitive to collaborate with.
Many OpenAI employees use GPT-6.2 Sol as their go-to model across technical and non-technical work.
But you are forgetting that Anthropic is going to have RSI next year. Claude will discover a new paradigm of Quantum Physics, transmute lead into silicon, and the hard takeoff will start. Only Anthropic and governments will be left, and that's only if they please Anthropic.
Announcing Generalised Superintelligence! (GSI) 🎉
GSI has achieved 250+ novel discoveries and breakthroughs across eight domains, including Quantum, Superconductivity, Chip Design and Mathematics through its Meta Solver Platform. GSI is extremely compute efficient compared to other frontier labs and excels at finding the hard structure in a problem first, before attempting it.
Everyone can contribute to scientific breakthroughs using the kind of superintelligence GSI is building.
Research preview out now at generalised.si
not picking fights with anyone but @theo is right when he point out that this is a dead end.
as models gets smarter, all of your efforts to FIX the harness are kinda moot.
the best agentic UX is just talking to the model. and the tools that facilitate you to do that across your computers is all you really need.
the best demos and products on this site are made by the folks who live by this minimal idea and Theo's team just gets it.
a lot of productivity promising hyperslop exists otherwise anyway which suits the past era neovim/emacs perfect setup dopamine hunter crowd iykyk.
@t3dotcodes is therefore winning.
apple seems to have waken up.
imo, apple has magnitudes of more potential than nvidia for getting this right.
their efficiency still sits in a different league.
Apple is developing 256 CPU cores and 640 GPU cores under Project Baltra 🗿🔥
•M6 (Komodo), M7 (Borneo), and M8 (Sotra) chips designed for remote Apple Intelligence processing and AI data centers.
These chips are designed for Apple's data centers in collaboration with Broadcom
soon.
clever compiler: "it can be proven that the pointer doesn't outlive the data, hence the check goes in the drain and your code runs faster"
watch this space.
trust nobody in the trenches.
except your friend. with the cheese. maybe.
EXFIL Rush: the first co-op game on @makeo_fun , built for the @exfildotfun community.
two rats 🐀
one king's cheese 🧀
five minutes to extract ⌛️
ps: you might have better luck with glasses on 😎
Every coin is a hill. We built the sewer underneath it. 🐀
Introducing EXFIL RUSH, one of our most ambitious games on Makeo, built for the @exfildotfun community.
Get cheese to the General before time runs out. The more you carry, the slower you move, and there's a black cat hunting you. Kill rivals and steal their cheese, raid the cheese vault, haul the King's Cheese with a friend, or use AI to bend the rules and forge wild weapons.
Go solo in a live multiplayer sewer, team up in co-op, or chase the jackpot leaderboard.
Live tomorrow @makeo_fun.
Play Your Way.
been experimenting with formal verification a lot lately.
something clever is cooking finally!
don’t know if this is peak psychosis but it wasn’t as hard as I thought to get where I’m at now with the current frontier models.
you should kill your CAPTCHAs.
for decades we built captchas to keep bots out.
but now the bots are your customers. people send AI agents to sign up, buy, and use your product for them.
block them and you block real business. let them all in and one person runs 10,000 free trials.
captchas ask "are you human?" the right question is "are you a good actor?"
so we built heedeo: let good agents in, keep bad ones out.
one free trial per person. not ten thousand.
heedeo.com :)