Tweet fijado
i'm 18 and hungry.
some people had change their life in a year.
i'll show you how (as i'm figuring out).
English
Sasha Girodon--Leclerc
186 posts
Sasha Girodon--Leclerc
@y_hungryguy
Chief Cloning Officer - building... fate deals the cards, and you play the game.
Paris Se unió Ocak 2026
251 Siguiendo17 Seguidores
At least I’ll have time to understand this tweet.
wafer@wafer_ai
🚨 OpenAI is launching GPT-5.6 Sol on Cerebras at up to 750 tps. Here's How kernels work on Cerebras Chips cerebras built a chip with 900,000 cores on a single silicon wafer. CSL (Cerebras Software Language) is a Zig-inspired DSL that gives you direct control over the Wafer-Scale Engine. the SDK is publicly documented and the programming model is genuinely the most alien one we've ever looked at. in CUDA you write one thread's perspective and launch millions. in CSL, there are no threads, no warps, no shared memory, and no kernel launches. you write code for individual Processing Elements: 900,000 independent cores arranged in a 2D mesh on a single silicon wafer. each PE has its own 48 KB of private SRAM, its own program counter, and a 5-port router connecting it to its 4 neighbors. that's actually it. no DRAM. no HBM. no cache hierarchy. 48 KB is your entire world per PE, where code and data must both fit. the programming model is dataflow. data moves between PEs as 32-bit messages called wavelets, traveling along virtual channels called colors. when a wavelet arrives at a PE on a specific color, it activates a task (a chunk of code bound to that color at compile time). tasks run to completion, then hardware picks the next activated task. tasks cannot call each other. they can only be activated. so instead of "launch N threads," you think of it like: "place code on PEs, define routes, let data flow." the memory model is also very different from GPUs. on an H100 you get 80 GB of HBM shared across all SMs. on the WSE-3, memory is 48 KB per PE, and there are 900,000 of them! this gives you 44 GB total on-chip SRAM at 21 PB/s aggregate bandwidth (vs 3 TB/s on H100). every access is single-cycle. no coalescing needed. no bank conflicts. no cache misses. but also no way to access another PE's memory. all inter-PE communication is explicit wavelet routing through the fabric. take a distributed GEMV for example. you would write a layout file that physically routes wavelets across the mesh. two PEs sit side by side. the left PE computes a partial result and routes it eastward. the right PE receives those wavelets from the west and accumulates. routing is defined at compile time. both operations are asynchronous and activate a task when they finish. you're physically routing data across silicon at 1 clock cycle per hop. cerebras is an extremely technically interesting & alien beast. reports 95-210x speedups over H100 on stencil computations. 3,000 tokens/sec inference on gpt-oss-120B. $10B+ deal with OpenAI for 750 MW of inference infrastructure. very exciting times for alternative accelerators! deep dive 1/6 by @gpuemi
English
I love when the war of the giants is a net positive for people.
Tibo@thsottiaux
To celebrate the launch of GPT-5.6 Sol, we will reset the rate limits again (twice) across ChatGPT Work and Codex over the next 24 hours. We want you to have the time to truly try ambitious tasks and get the hang of it. Happy exploring!
English
GPT-5.6 Sol is unbelievably good at creating and editing videos.
It can do motion design, product demos, and animations like this one I made by simply giving it a screen recording.
GPT 5.6 has the best design taste and significantly outperforms Fable, which relies heavily on repetitive design patterns.
To help you experiment with video editing on it, we just launched a collection of 100 ready-to-use skills that show what’s possible and help you get started with video editing using GPT-5.6.
These skills can create anything from motion graphics launch videos for your product to a 3B1B-style science explainer video.
You can also use them to edit existing videos: add captions, generate motion graphics, create voiceovers, redesign visual styles, translate into new languages, and much more.
If you want access to the full library, comment “VIDEO SKILLS” and I’ll share it with you. (You'll have to follow me so I can DM you.)
OpenAI@OpenAI
Sol, Terra, and Luna, our GPT‑5.6 family of models, are starting to roll out now in ChatGPT, Codex, and the API.
English
@mae_prina personally i use codex, limits are higher and sol is getting released tomorrow. the gap with fable may not be too big!
English
Mon neveu a 18 ans, c'est @mae_prina
Il est coincé dans une école d'informatique où on lui apprend du bullshit à longueur de journée.
Alors je lui ai dit une seule chose : arrête d'attendre qu'on te forme, apprends à build, absorbe tout ce que tu peux trouver sur internet, et ship.
Son objectif est simple. Sortir un SaaS qui lui génère du revenu d'ici la fin de l'année. À 18 ans, en partant de zéro, en apprenant tout seul.
Allez le suivre, c'est un vrai good guy.
Français
Wow: Claude can now make animated motion
Comment “motion” and I’ll send you the skill file
Romain Torres@rom1trs
I built a Claude skill that makes motion-style animation. > Paste a reference image > It generates frames with Nano Banana 2 > and animates the video with Seedance 2.0 Everything runs in Claude through the Arcads MCP Comment "Motion", and I'll send you the skill
English
@pato__project Les deux ne sont-ils pas très liés ?
Français
@natnaelgirma27 Listen to podcasts of @antinertia he is the goat, you will find your answers !
English
You’re building a startup from scratch. You can only hire ONE first growth guy.
A. Engineer who became « GTM Engineer »
B. Content creator → Growth
C. Ex-VC
D. Elite business school grad (with no xp)
E. Grinder with no prestigious background
F. Infopreneur turned app/saas builder
Who are you betting on and why?
English
@raph_guilhem Yeah a 200/mo claude max subscription, it’s the most cost efficient employee
English
Just discovered ultracode reasoning after months spent on codex.
just by hitting enter you can see your limit bar go to the red in real time. you now have 5 hours to go out and contemplate our beautiful world. I love it.
English
@raph_guilhem Happy birthday bro 🥳
This cake is definitely not ai
English
@vincent_maurin I’m more at the celio level net worth
English
happy to announce that I’ve reached Uniqlo level Net Worth
Usopp@Usoppu
3-Fig Net Worth: Uniqlo 4-Fig Net Worth: Uniqlo 5-Fig Net Worth: Uniqlo 6-Fig Net Worth: Uniqlo 7-Fig Net Worth: Uniqlo 8-Fig Net Worth: Uniqlo 9-Fig+ Net Worth: Uniqlo
English
Help me choose a thumbnail for our 0€ + 0% equity startup summer camp in Paris.
Which do you like most ?
Also there's 1 inspired by @parisbayarea poster from a hackathon, can you spot it ? 👀
English
In two months you can lock-in yourself and achieve something you always wanted, faster than you expected.
But at the same time, loose contact with a significant part of your friends.
Maybe Alex Karp was right on the impossibility of a social life in your 18s if you want to win.
English
Sasha Girodon--Leclerc retuiteado
Bro just discovered that he's been paying his API credits 45x the provider price (no shit sherlock)
But you're still wrong on one point though: the issue isn't that they're a wrapper, it's not even an insult tbf. Every tech company is a wrapper of something. Even model providers can be called wrappers.
The issue is in the value provided for that much of a multiple on credits pricing per unit: even on their middle plan (which only difference with the enterprise plan is credits volume in most companies), you get rate limited like crazy and still discover paywalls everywhere.
For instance you won't be able to launch five or six generations simultaneously, which is a crazy downside when you work at scale on video creation, whether it's ads or full movies.
I'm usually on the company side as most people are cheap and expect everything for free, particularly with AI models' pricing. To justify such an upsell your app must provide an insane amount of added value on top of the capacity of the existing models. I could have said that it's the case for their supercomputer, seeing how promising it looks, but even that seems only usable for a few minutes before hitting rate limits.
Good call telling everyone that doesn't know though, and it's crazy to think it still concerns SO many people
they had such crazy distribution that they were able to cover most of the call-outs on their scammy pricing
That said, personal preference but I use @fal instead of replicate for my flow inside Claude + Argil
Guillem@guillemcraft
higgsfield is LYING TO YOU it's true. they're an AI wrapper that charges you tons of credits for using the same API everyone else uses, and way more expensive what's the point in paying for monthly credits when you can just run Replicate and have the same results? i get it, they actually did a great job at wrapping things together and marketing it with tons of ads, but don't be fooled stop wasting money and add your github to Replicate and pay per generation instead (way way cheaper!) btw, i'm not sponsored by Replicate but i felt the urge to tell you, even more if you're trying to increase your MRR higgscam will steal all your money and you won't even notice i post about what works and what doesn't, and how to actually save money on marketing bc most of us don't have that much cash $$$ what do you think? did you run out of credits? will keep you updated
English