Daniel Carter

Artificial Intelligence Researcher specializing in Reinforcement Learning (RL).

[Daniel Carter @akalsk19]

@lingo24 Gpt line up is still much better even before they release the new models which would be very soon

[Daniel Carter @akalsk19]

@YurrrYurrr66926 @HarshithLucky3 I have said it once and I'll say it again: Gemini models should not be used for coding work. I have been using it for 30 minutes and I feel like punching my screen. I'm getting false flagged for absolutely no reason every two prompts even when I just say "hi" to it

[Daniel Carter @akalsk19]

@desa_th @HarshithLucky3 Now I'm getting this bug whenever the agent uses a skill.md file that it created. I can still use the model in different chats, wtf.

[Daniel Carter @akalsk19]

@HarshithLucky3 why is it more expensive than opus 5.5 on high and less efficient

[Daniel Carter @akalsk19]

@_mohansolo Hi sir, please fix the bug where I can't start a new chat for new projects. This is happening with the latest release , thank you

[Daniel Carter @akalsk19]

@lyraxana It's either benchmarks with no release or a release with no benchmarks.

[Daniel Carter @akalsk19]

@tonystark_mk3 @AI_Screening Be honest, do you actually use Claude? If not, yes, Gemini can do the job with some wrestling it's just not pleasant to use

[Daniel Carter @akalsk19]

@tonystark_mk3 @AI_Screening Cost: It's more expensive than Opus per task. Performance: Worse than Haiku in a real codebase. Speed: Much slower than Opus.
These benchmarks mean nothing when you try it in real work, not just one shotting useless 3D things
the only good thing about it is just chat and vision

[Daniel Carter @akalsk19]

@theo Seems like it's bigger than just cheating

[Daniel Carter @akalsk19]

@JamesDa55049936 Gemini will never be on Anthropic's level until Google replaces the entire board and runs it like a real, independent company instead of a hobby startup

[Daniel Carter @akalsk19]

@JamesDa55049936 If Gemini's top model (Carbon) is worse than Haiku at coding tasks, how is their cheaper, less intelligent model supposed to beat Sonnet?

[Daniel Carter @akalsk19]

@testingcatalog Google really invented a whole new kind of RSI: Retarded Self Improvement

[Daniel Carter @akalsk19]

@Mr_nexp Price doesn't mean anything if the model is retarded and spends so many tokens reading the same file 14 times

[Daniel Carter @akalsk19]

@ultrahumanite9 @imjustnewatai It's not a harness problem I used Claude in Antigravity too. Gemini just isn't trained for real coding work; it's only good for one-shots, and that's it.

[Daniel Carter @akalsk19]

@JamesMalsawm No, Haiku is better than Carbon and I'm not joking

[Daniel Carter @akalsk19]

@imjustnewatai then I gave the exact same prompt to Claude 5.5 Sonnet and it did it in like 20 seconds

[Daniel Carter @akalsk19]

@imjustnewatai Sorry for the toxicity, but whoever compares Gemini to any Claude model even Haiku in coding tasks is a retard. Like, it's not even close. Gemini is unusable in a real codebase; it's been reading files for like 10 minutes for a very small task that I even told it how to do

[Daniel Carter @akalsk19]

@LuminaBench Will it ever be better than Haiku? I highly doubt it