Installed @t3dotcodes to try Neon (after @theo raved about it) but couldn't get past the initial configuration hump. First I tried to add MCP servers, but T3 Code doesn't seem to have that in Settings. Then I started a session and told Neon to copy over all my existing MCP configurations from Claude Desktop (because I have a few dozen and don't want to do it one by one). Neon tried a lot of things, and after about 90 minutes of trial-and-error it still hasn't gotten it to work. It only sees the t3-code MCP.
I sure as hell won't be installing Antigravity just to try Neon. I guess I'll wait to see if Google makes it possible to use their models in other harnesses so that I can just hook it up to Claude Desktop.
I think T3 Code needs some way of seamlessly handling it for you. The experience should be: 1) Install 2) Connect your provider 3) Migrate all relevant settings from whatever harness you're currently using 4) Off to the races.
(I know T3 Code is "not a harness" but as a user trying to get things done I don't want to have to worry about the distinction. It should all be abstracted away.)
@bcantrill Yeah. I will say though, it is not surprising that so many tech leaders are siding with Jensen. It tells you a lot about what they value (growth, success, profits) and what they don't give two shits about (honesty, integrity, national security, following the law).
Tibo is so fucking damaged. OpenAI gives influencers early access and everything else they want so they glaze 6.2 Sol and Astra, and now even the influencers are hating on 6.2. so tibo writes a conspiracy theory about how they all tested an old version and nobody re-tested. your own glazers dont like the model and thats your explanation?? he is so deep in his bubble he cant accept that 6.2 is shit
TokenGremlin is one of the biggest grifters on ai twitter.
This guy spent weeks acting like he had extraordinary insight into gpt-6.2 astra. Claimed early access, posted supposedly exclusive videos and outputs without providing evidence, confidently predicted it would outperform fable 5.2, and repeatedly announced release dates that turned out to be wrong.
November 5, november 6, november 11. All wrong.
On november 9, he admitted he didn’t even have a reliable release date, only to confidently predict another one two days later.
He was also called out over the authenticity of generated outputs, and his response was to call gpt a cheater, claiming astra was secretly using image and video generation tools even when explicitly instructed not to. Yet he continued sharing supposed early-access material without providing evidence.
It’s genuinely astonishing how far you can get on this platform by sounding confident about things you supposedly know behind the scenes.
openai should have released nothing tbh
this is a disaster
it was frustrating to get no models from them, but we all thought they were cooking
turns out they were not cooking at all
one of the worst model drops ever
comparisons coming soon
@mattshumer_@ForwardEditor Yes, this matches my experience. This model (Astra) cannot work inside a mature codebase. Is very lazy and goes off on wild tangents too.
@daniel_mac8 Dan, it was always going to be this way. When OpenAI had no answer to Opus 5.5 for almost two months, it should not be surprise that they panicked, put a couple of models together hastily and fumbled.
@TheStalwart@jnconkle This actually turned out to not be the case. It has been more than six months since Fable came out and there are zero open models that can come close to its performance or intelligence.