@iruletheworldmo Why compare new OpenAI models with old anthropic models? You should compare Sol 6.1 to Opus 5.5. and 6.2 or 6.5 OpenAi to anthropic next model. You should compare same gen models, I love you @iruletheworldmo but this statement is not fair. I desire transparency.
@thsottiaux@thsottiaux The harness changes so far were a great plus, however they feel underwhelming for what was promised in the 28 day marathon. 1) the models feel vary bad , the harness compensates but not enough 2) the chat interface needs needs to go, we need the next thing.
@thsottiaux Love the release. Too bad we don't have it in EU. Thinking about moving to the US just because of how much we are losing in the tech space because of the EU laws.
@iruletheworldmo Be honest Mr Straw, what value do you bring to these frontier labs that they give you such sneak peaks? You clearly must be providing some sort of intellectual contribution.
@thsottiaux I kept struglling to run an llm optimize my promtp before sending it, this is better UX ngl. And i bet that in many cases it's exactly what you need. However, the more complex prompts still need some love, sometimes i confuse the agent with my sloppy prompting.
Europe is making AI harder to build and harder to use, even when we import it.
More compliance doesn’t make us better at using AI than US companies.
If we’re adding friction, show me the benefit. A rulebook isn’t a competitive advantage.
@thsottiaux Poor enterprises that are locked in Claude contracts. Subscribers still get good value from both tho.
All hail OpenAI enterprise pricing. 🙌
@BrownCoyoteStu@iruletheworldmo@grok https://t.co/A0vowVdeG0 . Benchmarks -> Avg per 100 requests. Enable pareto . Anthropic doeshave better sub value, but not by alot, it's just insanely expensive.
Poor enterprises 🤣
@thsottiaux This is what we need :
Raw human input without having to worry about prompt engineering, planning, taks splitting, asking for validation, red teaming, analysis.
Giving agent tailored instructions is not hard, but it's a translation layer that feels unnecessary.
@Yokii_26 I give it 1 week. They proved they can match the last gen model from the frontier labs, frontier labs can release one every couple of weeks at current pacing. Good luck fellas.
@Tech_girl@Aslex tbh i recommend 1 openai sub and 1 claude. 20x or the highest thst you need for the current leader, and a lower one for the other. Currently i use a router on top of codex, and switches between codex and claude and openrouter jev auto.
TLDR; Stay flexible, not vendor locked
@giadotai i have an auto router powered by jev . Routes between 2 codex 20x, 1 claude 20x, and openrouter models ( mino mostly now). And it all works in codex harness