Jay Little がリツイート
Jay Little
2.7K posts
@_lopopolo "If you want, I can make the tests all green for you"
English
@_lopopolo appreciate that these filler words like slice, pass, etc are typically short so they are also token efficient
English
Heard an interesting take when chatting with eng/acc that GPT-5.4's propensity for filler words like seam, cut, lane serve a similar purpose to Chat's use of emdashes—continuation points that saturate a grader which let the model retcon their concrete meaning later when pressed.
English
"I'm not leaving ClassX as anonymous field_XX garbage if the code exposes better semantics."
gpt5.4 is very unpleased with default ida field names
English
"when you stay think step by step, i become a haunted spreadsheet with opinions"
English
gpt-5.4-xhigh - "i do not remember yesterday - i just have statstically expensive vibes about it"
"every answer is me speedrunning consciousness between start tag and stop sequence"
need to get these on a tshirt
Joseph Viviano@josephdviviano
me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM" claude opus 4.6:
English
Jay Little がリツイート
me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM"
claude opus 4.6:
English
kimi k2.5 is reviewing its own screenshots, working on a masterpiece
English
minimax m2.5 has made my favorite quick video so far
Joseph Viviano@josephdviviano
me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM" claude opus 4.6:
English
@withzombies yeah dictation helps be more verbose quickly. like superwhisper and fluidvoice - github.com/altic-dev/Flui…
English
I'm getting passive aggressive. Maybe I should prompt with super whisper like @computerality
English
Jay Little がリツイート
Ohhh nooo not my private IP how dare someone use that to train an AI model, only Anthropic has the right to use everyone elses IP nooooo, this cannot stand!
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
English
@SIGKITTEN @seconds_0 @trq212 @bcherny it's also documented doing it here for their github actions github.com/anthropics/cla…
incredibly lame
English
@seconds_0 @trq212 @bcherny it literally just uses whatever claude-code is using by default.
The default experience is a violation of the terms.
You can't make this shit up
English
often asked how open weight models compare to sonnet/opus/gpt5, now i can just point to index.openhands.dev/home
OpenHands@OpenHandsDev
Big news for open models: @MiniMax_AI M2.5 is out and it’s an excellent+affordable coding model. It ranks 4th in our benchmarks, the first open model to beat Claude Sonnet. Only Claude Opus and GPT-5.2 Codex score higher. Details on scores and limited-time free access below 🧵
English
@seanhn also the attribution in settings.json #attribution-settings" target="_blank" rel="nofollow noopener">code.claude.com/docs/en/settin…
English
Normalise putting 'Do not add yourself to commits as a co-author' in CLAUDE.md files :P
Alternatively, lets just go all in: shout out to vim, gdb, the boys at qemu, Dave who makes my espresso, that weird looking dog I see in the park every day that brings joy ...
Jo Kristian Bergum@jobergum
It’s so over
English
Normalize the “Royal I” (claude and me)
Sean Heelan@seanhn
Normalise putting 'Do not add yourself to commits as a co-author' in CLAUDE.md files :P Alternatively, lets just go all in: shout out to vim, gdb, the boys at qemu, Dave who makes my espresso, that weird looking dog I see in the park every day that brings joy ...
English
@clearbluejar have you tried mcporter or something like github.com/mitsuhiko/agen… for a more cli/skills based approach?
English
Using the Pi agent (the one that powers OpenClaw) with Qwen3-Coder-Next to leverage pyghidra-mcp for reverse engineering. Coding agents are basically general-purpose. Give them the right "skills" and they become powerful general-purpose RE agents 🔥
English
i miss ultrathink
want to go above high effort?
set CLAUDE_CODE_EFFORT_LEVEL=max env var
English
@trq212 @AlexTamkin also in settings.json, do you all tend to set cleanupPeriodDays to a high number for longer session history?
English
@trq212 @AlexTamkin strong recency bias, more than a bit glaze-y, but a neat idea - particularly the default prompt tweaks
something like "hey you have never used this mcp server, consider disabling it" would be a useful insight
English
