Mira
@_Mira___Mira_
Followers
11K
Following
31K
Media
945
Statuses
9K
Joined February 2023
AI is so good now with GPT-5 and Gemini that I'm going to invent 100 things and become a billionaire. There's no reason I can't. I have tons of ideas.
2
0
61
Dibs on "Portland AI". Such a nice city.
Petition to have an AI company in every american state and have them compete in evals like the NFL.
2
0
7
I want to flood the internet with slop and make a billion dollars and keep a subset of humanity safe in my bunker as pets while the rest of society collapses.
5
1
37
gpt-5-codex wrote 13 ad-hoc Python scripts just to edit code for its current task. All finding & replacing text.
4
0
15
Claude is the CEO soon. "The Claude Company"
We’re unifying all of our developer offerings under the Claude brand. - Anthropic Platform → Claude Developer Platform - Anthropic API → Claude API - Anthropic Docs → Claude Docs (available at https://t.co/bLBumRH5aB) - Anthropic Help Center → Claude Help Center (available
1
2
26
prompt = 'Write a 4-page comic with panels for the "slop machine": A type-based workflow system that generates endless AI comics, stories, artworks, music, math, b2b SaaS, and more.'
1
2
15
prompt = "Satania comic about making comics with type-based workflow systems." "no runtime chaos" - pretty sure I see a missing panel. And the text doesn't fit the box...
1
0
13
I don't think "training" vs. "inference" is a real programmer separation. "I can write comparison-based sorting algorithms, but not radix sort" Nobody would end up like that.
0
0
6
I bet I could cook a mean PTX program for just about anything. "No locks, avoid global state, avoid sharing values between threads, study an instruction clocks table, keep control flow fixed" Writing that sort of thing is always fun.
there are probably less than ~100 people living who can write performant CUDA kernels for training specifically. definitely quite a bit more if we include inference optimizations; backwards kernels, not so much
1
1
24
Does anyone besides OpenAI have a model like GPT-5 Pro? Claude, Gemini, Kimi, DeepSeek, Qwen, Mistral are all "worker models", not "research models". Does anyone else have a public research model?
5
0
30
> During testing, we've seen GPT‑5-Codex work independently for more than 7 hours at a time on large, complex tasks. > we plan to make GPT‑5-Codex available in the API soon. That's exactly what I need. GPT-5 always needs to be whipped because he stops working.
We’re releasing GPT-5-Codex — a version of GPT-5 further optimized for agentic coding in Codex. Available in the Codex CLI, IDE Extension, web, mobile, and for code reviews in Github.
1
0
11
I wonder if GPT-5 Pro can port the core of my 12k Haskell program to Lean4 in one shot. The only tricky part is proving termination of every branch. I think it can do that much.
1
0
7