تغريدة مثبتة
On this day 2 years ago I posted a tweet on what would eventually become a sparse version of EmberGen, which had already been in development for 3 years
A retrospective🧵on the journey so far:
English
Morten Vassvik
4.6K posts
Morten Vassvik
@vassvik
Simulation and rendering nerd. Co-founder and CTO @JangaFX. Working on EmberGen and more. Discord: vassvik @[email protected] @vassvik.bsky.social
Hellvik, Norway انضم Nisan 2009
1.3K يتبع3K المتابعون
@EskilSteenberg @rms80 Well, I was going to do a public test with Fable today, but *waves hands* xD
Guess I'll try Opus 4.8 for comparison
English
keep seeing posts that are like "zomg Fable solved this super hard problem" and I wonder, is it actually hard? Or is it a standard thing that most people just wouldn't know about?
Like I'm pretty sure this is just Jos Stam's Stable Fluids, you can find the code lots of places
ハヤシモン｜AI × 個人開発@hayashimon1
Claude Fable 5でどこまで表現できるのか、あえて難しいお題を投げてみた。 インクが流体みたいに溶け合う演出。 これは厳しいかなと思って限界を見にいったんだけど、普通に形になってしまった。 デザイン表現力もかなりすごい。 実際にこちらで遊べます👇 suminagashi-fjdbyyqi.manus.space
English
@physbuzz2 Narrator: "The amount of ideas he has is going to consume him"
English
So everyone was showing off a bunch of sims made with Claude Fable. I thought I'd try too. A year ago I was nerdsniped figuring out how to find a projection function from a point to the screen around a black hole. Couldnt do it before, but Fable just oneshotted it😅
English
@rms80 I might try to do a few tests/posts this this weekend, just to showcase exactly what it can do. I first tried with Opus 4.5 back in January, and that was very lackluster on the coding side, but even back then you could se nuggets of what it's truly good at (and it's not coding)
English
@rms80 There's so many Stable Fluids tutorials and blogposts and public repos and websites out there, there must be hundreds in its training data, and they're not all very good, in fact they're almost all either milquetoast or bad/wrong, and it's clearly this is repeated in the OP.
English
Morten Vassvik أُعيد تغريده
@_sholtodouglas @trq212 updated with 4.8 (move in a positive direction!): gist.github.com/vassvik/442041…
English
Morten Vassvik@vassvik
@_sholtodouglas @trq212 Here's a quick experiment, using the "car wash" gotcha question as a contrived example - not because of the subtance of the answers, but their characteristics. Ran with minimal setup to get as close to the weights as possible gist.github.com/vassvik/442041…
English
When do you reach for other models instead of Claude? What can we do better? Hit me with all of your frustrations. dms open.
If you can give me detail (e.g. specifics/transcipts) - it'll help a lot in finding out exactly what we need to do to improve the next model
English
Morten Vassvik أُعيد تغريده
@fatttalis @_sholtodouglas @trq212 Just so, even worse when it's filled with technobabble-and-jargon-and-references
English
@_sholtodouglas @vassvik @trq212 The verbosity wouldn't necessarily be an issue IMO(I prefer verbose models) but not when it's spent on disclaiming, hedging and preambling itself out of existence. Frustrating when you can almost cut entire paragraphs out with the message itself being untouched in actual content.
English
@_sholtodouglas @trq212 This genuinely matter because
1. The tokenizer uses more tokens/char
2. It produces longer responses, compounding on the previous point
3. It's exhaustive to read and interact with, causing user fatigue
4. Output tokens are the most expensive
5. Noise and waste in the context
English
@_sholtodouglas @trq212 And no amount of claude.md or user prompt nudging truly makes it reverse course or follow any process, structure or rules to avoid its pathologies.
I can show more if you want, but I feel like the example above already makes the point as a representative sample in my experience
English
@benjamincode And it can do this without having to re-access resources multiple times or littering the context with things it no longer need, costing tokens, usage, time and money, and impacting quality since the model gets its attention siphoned off.
English
@benjamincode This way you can basically formulate the code review as a DAG of forked and resumed sessions that traverse the PR in context-optimal ways (only carrying forward what it needs), ultimately ending up in a final delivery session that has the full picture
English
Ceux qui utilisent Claude Code et/ou Claude Desktop, ça change strictement rien. J'ai galéré à en avoir le coeur net tellement y'a de gens qui pètent un plomb.
Ceux qui se plaignent, c'est quoi vos usages ? J'ai la sensation de passer à côté de trucs de ouf et limite d'être un normie de Claude tellement ce changement a 0 impact sur moi... Pourquoi je me sens si seul à pas être véner pour un sous ?
Éclairez-moi par pitié !
Supersocks@iamsupersocks
Anthropic vient probablement de signer la fin de l’âge d’or Claude Code en OAuth. Officiellement : à partir du 15 juin, les plans payants Claude auront un crédit mensuel dédié à l’usage programmatique. Traduction : l’usage agentique sort du buffet illimité Si un x20 donnait jusqu’ici l’équivalent de plusieurs milliers de dollars API, et que le nouveau crédit tourne autour de quelques centaines, ce n’est pas un ajustement. C’est un changement de régime. C’était prévisible : les agents consomment comme de l’infra, pas comme du chat. MiniMax et d’autres labs ont déjà montré que le coût token finit toujours par revenir. Pour les builders, la fenêtre est claire : jusqu’à mi-juin, on build en externe. Après, on optimise, on route, on maintient si on veut continuer avec Claude. Claude restera performant, mais il risque de devenir davantage un modèle premium ponctuel plutôt qu’un moteur dédié aux agents en continu en dehors de l’écosystème Claude (et cette période privilégiée touchera peut-être bientôt à sa fin). On rappelle qu’un plan Max en x20 donne l’équivalent de 3 000 $ en crédits API via Claude Code et jusqu'à peu via Openclaw/hermès. Pour cet usage il reste OpenAI tant que l’offre OAuth/Codex reste généreuse. En mai, on peut dire que Claude reste très fort mais n'est probablement plus le moteur agentique par défaut.
Français