Will Stone

ai lead @multisynq

[Will Stone @tribixbite]

Winter is the best time to build 🔥
Alpha footage of EIGHT of our *completed* spaces:
• Future Dome
• Pirate Dome
• King Dome
• Hell Dome
• Forge Dome
• Samurai Dome
• Explorer Dome
@Imaginarium_SOL's Hub

What are you building?

#BullishOnBuilders #SolanaNFT #metaversol

[Moonshine Labs - DM for Collabs @MoonshineLabs]

Are you ready to be a builder? Are you ready to explore the metaverse? With metaversol the possibilities are endless. #solana #bullishonbuilders

[Will Stone @tribixbite]

@bcherny Sweet, maybe I'll try it if I ever use my desktop again. Meanwhile, I used cc to build a new keyboard just to make it easier to use cc on mobile (in termux).
Check it out here: https://t.co/K7gaKI7pBW
2,414 commits so far heh

[Will Stone @tribixbite]

@docfork_ai where do you pull libraries from? I'm trying to access docs for @multisynq/client npm. https://t.co/0lyTBXhmQB (hosted on github at multisynq/docs)

[Will Stone @tribixbite]

@theo why are you using eslint what is this 2023?
biomejs

[Will Stone @tribixbite]

https://t.co/wvco7ie7Ee

[Will Stone @tribixbite]

Claude artifacts has been broken for months. New and npm@latest imports 404 bc cdnjs' update pipeline was turned off in March. Last week, @multisynq politely asked @cloudflare to fix it. As of 6/16/2025, all cdnjs packages have been updated to their latest versions. ⬇️

[Will Stone @tribixbite]

@alexalbert__ Too bad cdnjs has been broken for 8+ months making it impossible to import the latest version of most packages or newly added packages.
https://t.co/9PQKDHBmbR

[Thomas Wolf @Thom_Wolf]

Wow - open-source LLMs for code are so back

Just one week after the latest DeepSeek release which brought vibe-coding for free, here is the high profile @allhands_ai team dropping a 32B model with matching performance on software engineering tasks benchmarks like SWE-Bench Verified. This is 20 times smaller than the DeepSeek model!

Results obtained by post-training with reinforcement learning in a code environment: remember the 2025 trend of reasoning models in math/code/etc trained with RL - yep, it strikes again

Feels like the open-source world has started to crack the secret code generation sauce of closed models like Claude Sonnet.

And we're still very much at the beginning, this is so exciting!

[Will Stone @tribixbite]

@ManyThorns @i0ns0 @CodeFryingPan mind linking any good articles about this, if you have any handy? curious

[Will Stone @tribixbite]

@CodeFryingPan alright so what did it miss and can you defend your claim about 'most active open source community'? not trying to attack/accuse, strictly curious what your criteria was.

[Will Stone @tribixbite]

the best-in-class projects tend to give a feature/metrics table comparison~ this would be particularly helpful and effective to back up that claim ('most active')
there are a lot of open source ai tools out there. it can be hard to pick one without feeling like you are informed/aware of alternatives. so while telling people about your competitors may seem counterintuitive, it's actually a great user acquisition strategy

[Will Stone @tribixbite]

@elder_plinius @JacquesThibs this is as far as I got. 139 seconds thinking time. didn't know about the rate limit, was just playing around trying to flip alignment polarity

[Will Stone @tribixbite]

@adlonymous https://t.co/NaR25QoE8z is the only one that seems legit. i've run a worker for it, things seem to be functional which im not sure can be said about any others

[Will Stone @tribixbite]

@darkmodeprince @MeritByFrankie Lol yw https://t.co/H69wg8Ga30

[Aadil (mous) @adlonymous]

1/x
Solana is cheap and fast by virtue of it's 8 key technical innovations, one of which is called Gulf Stream, but what even is it?

(More technical article next week)

Let's go 🧵👇

[Axel Peytavin @ax_pey]

We gave a tiny body to gpt-4o

Fully autonomous, give it just one prompt

Robot sees -> thinks -> does -> repeat

Can also:
• pick up trash
• shake hands
• talk with you

All for $200 off-the-shelf

Building a bigger version
What should we have it do?

[Thomas Wolf @Thom_Wolf]

This might be one of the most important 45-mn read you could indulge in today if you want to understand the secret behind high performance large language models like Llama3, GPT-4 or Mixtral

Inspired by the @distillpub interactive graphics papers, we settled to write the most extensive, enjoyable and in-depth tech report we could draft on the science of creating high quality web-scale datasets, detailing all the steps and learnings that came in our recent 15 trillion tokens 🍷FineWeb release

And it's not all, in this article we also introduce 📚FineWeb-Edu a filtered subset of Common Crawl with 1.3T tokens containing only web pages with very high educational content. Up to our knowledge, FineWeb-Edu out-performs all openly release web-scale datasets by a significant margin on knowledge- and reasoning-intensive benchmarks like MMLU, ARC, and OpenBookQA

To close, we make a number of surprising observations on the "quality" of the internet it-self which may challenge some of the general assumptions on web data (not saying more, I'll let you draw your conclusions ;)

Take a walk through it, you won't be disappointed 👇

[Will Stone @tribixbite]

@bilalghalib Hah you've got some great ones. I think we have similar tld taste... If I didn't already have too many I'd ask for thisbutlocal and maybe half the prompt ones, but I'd love to give bloom.api to my friend J. Bloom who said two days ago he wanted a short domain w/his last name...

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

Reverse Turing Test: AI NPCs try to figure out who, among them, is the human

Aristotle is GPT4
Mozart is Claude 3 Opus
Da Vinci is Llama 3
Cleopatra is Gemini Pro

The funniest part?

3 of the 4 models guessed correctly... because the human's response was too dumb 😂😂😂

[Trung Phan @TrungTPhan]

Google AI overview fiasco, explained

[Will Stone @tribixbite]

@mattsilv @web3pm @Gribbzzzz @moonbirds @yugalabs @kevinrose @discord A protection mechanism would lead to (vanity invite...) squatting, probably best to maintain the discord invite as an owned (sub)domain or path param redirect from project site. 0 reason to use discord's hosted vanity invite system when you own your own domain imo