The problem I have with this is that you can't responsibly claim at a scale like yours that a 20% (!) layoff is explained by AI. Let's just take non-coding, you're saying CF has (or even will have) agents that can reliably do all of that work? If you're not earnest, then you're paving the way for more people outside of industry to blame AI mistakenly which will have second-order consequences.
Same thoughts as Corey. I've used it every day since it has come out. It writes in this very terse and almost cryptic way. It also always uses different words for concepts I've already used very particular words for in a session. This was widely discussed for 5.2 online, whip smart (just like 5.3-codex) but writing that feels very alien like and non-coherent/clear.
I still remember (back in oct lol) when @steipete was the biggest claude hater on the planet. Even accused of being a paid shill for codex. And now opus is the defacto agent of choice for openclaw, even though he still hates it for coding. Crazy how fast things are moving!
I still remember (back in oct lol) when @steipete was the biggest claude hater on the planet. Even accused of being a paid shill for codex. And now opus is the defacto agent of choice for openclaw, even though he still hates it for coding. Crazy how fast things are moving!
There is such an annoying intelligence gap between the text and speech models that I just wish google/openai would give you an option to use voice mode the old way with stt/tts. I'd take some additional latency for it any day of the week over interruptibility/responsiveness.
PSA if you use perplexity search just jump ship over to @grok 4.1 thinking right now. the responses actually stack up against its sources, perplexity still lies half the time. I've instructed it to use x posts and it finds the most relevant stuff almost every time.
@james_almeida@venturetwins Nah itโs the speech models sucking at instruction following. The model should listen to her as soon she says โdonโt say anything except your lineโ. The speech native models are clearly not properly trained on instruction following like the chat models are.