Yeah, yeah, you're an early-twentysomething new grad with a chip on your shoulder, not taking anything too seriously. Been there, done that, still deal with your kind at work all of the time. Bookmark it for either 10 years down the line or some random time when you're high enough to actually think as deeply about things as you should.

Coding-related use-cases are only 25% of the methodology of their updated Intelligence Index.

Because you're subconsciously evaluating it by their "coding index" rather than by their "intelligence index". artificialanalysis.ai/agents…

So: The reason the model is doing a shitty job is because your prompts very obviously suck. But that's a good thing. The fact that it's your fault means that it's within your power to fix it. The only things you can ever improve are the things that are your fault.

Something maybe worth reflecting on: Responsibility and power are literally different words for the exact same thing. The only things you can ever improve in your life are the things you can accept the blame for. Always finding a way to blame others leaves you helpless.

The longer you fail to take responsibilty for improving your communication in your prompts, the more you'll needlessly suffer from "the model being dumb" when the communication issues are clearly on your end.

Gap between TerminalBench and FrontierCode is basically a "read my mind and correctly guess what my underspecified 7-word prompt means" benchmark. No surprise that Opus leads in that; Anthropic has geared their model more toward understanding people over systems for a while.

The reason it's "completely changing the design" is because you weren't specific enough about what you wanted. The reason it couldn't tell what tab you're on is that you weren't specific enough to explain what to read and how. It's all just failures to communicate properly.

Just a bunch of <= 7-word underspecified "read my mind please" types of prompts.

Seems like a theme.

Still cooking on a DeepSeek V4.1 Flash recipe for 4x DGX Sparks. Pretty significant wins on single-stream decode compared to competitors, and absolutely dominates in prefill.

It's an index, not a benchmark lol. The only thing they've done is change the index tonight. The models didn't suddenly get any dumber just because they score lower on this particular index. The new addition to the index is the "frontier STEM" tests that frontier models only just started breaking past single-digit scores in during the last release cycle.

People who start fights and get themselves all upset enough to block you... lol. I don't understand why these people even bother starting arguments and upsetting themselves so much. Created their own problem.

I believe that you genuinely believe that.

People who call out shitty AI models are "pathetic"? I'm not even going to pretend that you have any reasoning for saying that. You don't lol.

Also, there is no need to be personally offended on Liquid AI's behalf lol. It's not as though you made the models; There's no reason to have your ego all intertwined with the concrete fact that they just suck really bad.

That claim isn't even close to relevant to anything I said or did lmao.

Yes, I do say so. Thank you.

Most of the open-weight losses come at the hands of frontier STEM weakness rather than everyday coding and similar work. It makes sense. They're more like 110 IQ office workers rather than 140+ IQ PhD rocket scientists.

This is the methodology btw.