rami

AI/ML engineer building agents, evals & strange interactive tools. Shipping in public.

[rami @vicilah]

Benchmarks aren’t even real anymore.

The labs making these LLMs are overfitting the model weights to the benchmark dataset.

The ONLY way i personally believe testing a model is seeing its capabilities across real world tasks like UI, or debugging a specific problem.

[orcus108 @orcus108]

WHAT THE HELL is happening in AI?

A 3B parameter model just put up coding benchmark scores in the same league as Claude Opus 4.5.

3 BILLION.

The weights are on Hugging Face, anyone can test it.

I genuinely don't know if this is a breakthrough or if the benchmarks are broken.

[rami @vicilah]

@kimmonismus not to mention they also limit their models on the chat interface, whereas i’m with 5.5 speaking to it without being conservative of my prompts.

[rami @vicilah]

@kimmonismus why do you think claude are surpassing openai in growth? I tried claude code for fable when it came out but even when i was using a older llm like opus 4.7 my limit would hit way faster than codex.

[rami @vicilah]

@Youssofal_ I like where the world is going, curious to see when open weights beat frontier. Also happy birthday @Youssofal_

[rami @vicilah]

@theo yeah haha, I feel like so many people on the ai train hype anything that’s packaged as “groundbreaking”. Like fusion style model panels for example, cool benchmark but cost/latency doesn't make quite sense. That's the part a lot of non-technical people miss.

[rami @vicilah]

@beka_saparbek TaylorCV, an app that tailors your cv to the job posting.

[rami @vicilah]

@kimmonismus Ah great insight thanks chubby :)

[rami @vicilah]

Interesting idea, but I’m trying to understand the economics here.

If Fusion is running multiple models in parallel + a judge/synthesizer, shouldn’t token cost scale pretty aggressively with panel size? @OpenRouter

[OpenRouter @OpenRouter]

Introducing the Fusion API, the smartest compound model in the market.

Fusion achieves Fable-level intelligence at half the price.

How it works 👇

[rami @vicilah]

Very intriguing

[AMD @AMD]

AMD Ryzen AI Halo. The ultimate local AI developer platform.

Pre-order now: https://t.co/Ny0ZV8LOYi

⚡ Up to 128GB unified memory
⚡ Support for models up to 200B parameters
⚡ Windows & Linux support
⚡ Ready-to-run AI workflows out of the box

Build, prototype, and deploy locally without cloud constraints.

[rami @vicilah]

Fable 5 rebuilt a 150 year old mechanical computer from one prompt.

it actually works

try it

https://t.co/KuMvc10G4l

[rami @vicilah]

@MatthewBerman frontier is temporary

the real pressure is open weights getting “good enough” faster than closed labs can widen the gap

[rami @vicilah]

frontier models are slowly turning into metered compute

Use them while you still can

[Theo - t3.gg @theo]

10 days left to escape the permanent underclass

[rami @vicilah]

@memefomo This is great, def using this!

[rami @vicilah]

@XScharo Something I don’t have sadly😔

[rami @vicilah]

@ihateoop Nice selfie, didn’t know you’re black

[rami @vicilah]

@pullupso Hey nigga

[rami @vicilah]

@MaddieGPT @BCRT @finnbags this was made before by someone else. this is a larp.

[rami @vicilah]

@earlTrades $ijustfoundsomethingcrazy

[rami @vicilah]

@ihateoop I used u as my EL just now pls don’t be mad