Sabitlenmiş Tweet
bling
4.2K posts
bling
@blingdivinity
Artist, Open to Interpretation 𖣘 American Undergraduate 𓉱 Casting Mathemagical Spells 𝜆
East North Central Katılım Mart 2020
2.2K Takip Edilen1.1K Takipçiler
during peak ZIRP era it was widely reported that for every $1 VC's invested in startups, 40¢ went to Google, Meta, and Amazon for ads or cloud services.
i wonder what the figure is now for agentic coding tools from OpenAI/Anthropic?
its gotta be brutal. your startup feeds claude
English
if your introspection is making you think more, it's not introspection, it's rumination
good number of ruminators falsely label themselves deeply introspective
but contrary to introspection which helps metabolize thoughts/perceptions & lead to action, rumination only ends up digging a deeper hole, and prevents action, because preventing action is the point!
Nick@nickcammarata
i think this meme is hilarious. my take on all this: the point of introspection is to end up thinking less, not more, to be more in the flow, more productive, to dissolve into being itself. if your introspection is making you think more i recommend getting another one
English
OpenAI's industrial scale CoT monitoring system would be ten times harder to do with o3 style neuralese. that is why gpt-5.4 CoT is in plain english. sorry thinkish bros, we must tie down our synthetic intelligence giants
Micah Carroll@MicahCarroll
Today we're sharing how our internal misalignment monitoring works at OpenAI – great work by @Marcus_J_W! 1. We monitor 99.9% of all internal coding agent traffic 2. We use frontier models for detection /w CoT access 3. No signs of scheming yet, but detect other misbehavior
English
interesting to see o4-mini beating 5.2 in nonagentic setups
no free lunch theorem is real
Lossfunk@lossfunk
4/ We tested GPT-5.2, O4-mini, Gemini 3 Pro, Qwen3-235B, and Kimi K2 across 5 prompting strategies. Models scoring 85-95% on HumanEval scored 0-11% on equivalent esoteric tasks. And every model, every language, every strategy scored 0% beyond the Easy tier. Not 2%. Not 5%. Zero.
English
@CFGeek @gwern idk looks like the post-COCONUT latent reasoning research is making progress. although having some amount discrete structure seems to be an essential engineering component
arxiv.org/pdf/2510.07745
arxiv.org/abs/2509.20317
arxiv.org/abs/2502.05171
English
@blingdivinity @gwern I continue to be a neuralese disbeliever but a Thinkish believer
English
@loss_gobbler certainly. mind reading is simultaneously the best tool for defending against a possibly adversarial intelligence, but also never definite, never enough. superintelligence is a dangerous game
x.com/blingdivinity/…
bling@blingdivinity
Yes i totally see this and will probably work but there are no guarantees that certain intel level it wont find a code to communicate/think internal ideas in on an abstraction dimensionality that the evaluator didnt know existed. Just like animals think ur outward expressions are a reliable indicator of intent. Whoops there was a whole extra level now im dead
English
@blingdivinity I think about this a lot. it’s probably net true, but it could easily backfire.
there’s a scenario where the “self programming/non-english representations” remain or grow, just pushed into more subtle encodings.
eg little grammar motifs, keyphrases, etc
English
you're a researcher at openai trying to form a vision of how artificial general intelligence can simultaneously coexist with humans while it radically transforms our world and what you come up with is *that it could help convince uber drivers to save for retirement*???
Houda Nait El Barj@Houda_nait
English
good interpretation of the results. but overall pretty useless benchmark. there are plenty of interesting and actually useful esoteric langs that couldve been used that have unfamiliar semantics, not just difficult syntax. not sure this is really measuring OOD reasoning: transpiling is mentally taxing to do within the weights, but lesso novel problem solving.
English
Everyone please read the whole thread. It can be simultaneously true that:
A) Much of the current coding capability is stored in the model weights as “memorization”**
B) The models ALSO have slower, general reasoning capabilities for OOD contexts. System 1 vs. 2 thinking.
Lossfunk@lossfunk
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
English
but how long will this equilibrium of natural language CoT last? if a lab can get full-neuralese-thought-vector style chain of thought working, it will be a huge advantage: the higher bits of information per step will make the models faster and smarter, but interpretability will suffer.
English
the key when talking about right-leaning is to specify “to the right of what?”
i hold classic anglo stark trek liberal ideals, but you would have to deport millions of people to make that happen. or at least force the breakup of ethnic enclaves Singapore style. liberalism is right leaning of status quo woke
English
high-openness right-leaning is just called being extremely tribal. you're functionally an orientalist for the fine china and exotic spices of cultures that produce works of art that your spiritually gelded one cannot.
alco ⊢ ꙮ@qualiascript
being right-leaning and high openness is so funny. "this is one of my favorite musicians, i disagree with everything they stand for, highly recommend"
English
@memeticsisyphus or some of us are not so emotional stable and we do care, but try to be true to our beliefs anyway. key word is try
English
Yes we pair it with emotional stability so we don’t care if someone doesn’t like us back.
alco ⊢ ꙮ@qualiascript
being right-leaning and high openness is so funny. "this is one of my favorite musicians, i disagree with everything they stand for, highly recommend"
English
@trishaepan @eugenewei its the word for jew in german.
anyway "yo semite" might make this best one haha
English
My friend Kevin told me about this page of Unparalleled Misalignments by Ricki Heicklen and it will be one of those web pages I turn to for years and years and just chuckle with deep pleasure.
TED Talk —> Edward Said is just 🤌🏼
rickiheicklen.com/unparalleled-m…
English
@animalologist ADA would def classify this as a disability so your job is safe too. well as long as youre not customer facing
English
There may have been a time when the idea of being randomly nude would petrify me, but we’re well past that. This sounds hilarious. Lemme stress somebody out. Or gaslight someone. No problem. Six figures for a superpower I can wield slightly maliciously? Yeah I’d take that easily
Real Post Folder@RealPostFolder
English
