aryaan ✨ (@halfCircuit_) header
aryaan ✨ (@halfCircuit_) profile picture

serial chiller

UY Scuti
Joined June 2012
Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)Photo by aryaan ✨ (@halfCircuit_)
Half my feed this week is "AI just solved [impossible thing]". P vs NP one day, a GPT-3 moment for robotics the next. Meanwhile the real work is still someone spending a week getting a model to do one boring task reliably, without hallucinating. The demos are exponential. The reliability is not. What's the gap you keep hitting?
297
If you had unlimited budget, what is the one thing you can invent with your current skills that would benefit humanity?
1
223
Weightlifting is the most honest sport. The bar doesn't care who you are.
1
210
Spent the weekend making my training data worse on purpose. The model was too clean. It had never seen a typo or someone cramming 'ls -la' into 'ls-la', so it fell apart the moment a real human touched it. I added 500 messy, realistic examples by hand. Accuracy on real input jumped more than any architecture change had. Your model only handles inputs as ugly as the ones you trained on. What's the messiest data that ever saved your model?
1
2
222
most 'agent skills' this week are just a prompt with extra steps.
2
1
167
My eval was lying to me for a week. The model looked great because my test set leaked from training. Every "win" was just memorization. I rebuilt it by hand and the real accuracy was ugly. The eval was the whole project, and I built it last. What's the sneakiest bug that's fooled you?
2
2
185
aryaan ✨ retweeted
Replying to @Teslarati
The rate of improvement from original GPT to GPT-3 is impressive. If this rate of improvement continues, GPT-5 or 6 could be indistinguishable from the smartest humans. Just my opinion, not an endorsement. I left OpenAI 2 to 3 years ago. Am a neutral outsider at this point.
243
1,050
595
9,813
You can't blocklist your way to sounding human. Ban 'delve' and it writes 'explore in depth.' Ban em dashes and it finds the semicolon. A banned-words file just teaches the model synonyms. Real voice needs something specific to imitate. What's worked for you? nitter.gq/Voxyz_ai/status/207885…
𝘀𝘁𝗼𝗽 telling Claude Code/Codex "no em dashes". 𝘀𝘁𝗼𝗽 telling Claude Code/Codex "stop saying delve". 𝘀𝘁𝗼𝗽 telling Claude Code/Codex "don't sound like AI". you never gave it a 𝘄𝗿𝗶𝘁𝗶𝗻𝗴 𝘀𝘆𝘀𝘁𝗲𝗺. every README, PR description and landing page ships in the same AI voice, and you ban words one at a time. orwell wrote the fix 𝟴𝟬 𝘆𝗲𝗮𝗿𝘀 𝗮𝗴𝗼. six rules, 1946. paste them into your global CLAUDE.md / AGENTS.md and every session picks them up: → never use a long word where a short one will do → if it is possible to cut a word out, cut it out → never use the passive where you can use the active 6 blocks you can copy-paste directly 👇
1
192
True at the frontier. Also true, most real-world tasks never touch it. I'm training a 12M-param model on a ₹30k budget to turn English into shell commands. For one narrow job, small and cheap wins. Not everything needs an H100. nitter.gq/nicbstme/status/207891…
Guys, the world is so compute constrained. I honestly don’t understand how anyone can think otherwise. Almost nobody is seriously using AI today, and we’re already hitting capacity. Everything is a bottleneck: land for data centers, permits, electricity, grid connections, chips, construction, cooling. Demand is growing far faster than supply, and expanding supply is insanely expensive. In this game, the best-positioned companies are the ones with massive net income from non-AI products. They can fund hundreds of billions in CapEx while everyone else has to raise money just to stay in the race… and remember if you are not at the frontier you have no pricing power so you won’t be able to fund your compute need.
90
Swapping out 'delve' and 'tapestry' is ~20% of the fix. What survives every word-swap is cadence. Sentences all one length, points in neat threes, a question answered in the next breath. Fixing the rhythm is the real work. Grabbing this. nitter.gq/petergyang/status/2079…
I’m sick of reading AI slop, so today I’m open-sourcing my /no-ai-slop skill that removes 20+ slop patterns from any piece of writing. 📌 Get the free skill here: github.com/petergyang/no-ai-… If you find it useful, please consider starring the repo so more people can find it. Why I built the skill: I use AI to edit my writing because it helps me fix spelling, grammar, and clarity. But even the best models keep producing the same slop that this skill removes: → Binary contrasts: “It’s not X. It’s Y.” → Throat-clearing openers: “Here’s what nobody tells you.” → Fake-profound endings: “The future isn’t coming. It’s already here.” Use this skill responsibly. I always write a first draft manually before iterating with AI on edits and I make sure to do another manual pass at the end as well. That’s in contrast to using AI to automate pumping out slop end-to-end. 📌 Read my full post for more on how I try to use AI responsibly to edit without giving into the dark side: creatoreconomy.so/p/use-my-n…
52
aryaan ✨ retweeted
Introducing Willow. The world's most accurate voice dictation tool. Willow learns from every correction you make. Automatically. Auto-learning plus context-awareness unlocks 99% word accuracy. No other dictation app comes close. RT + comment "Willow" to get a month free.
117
61
48
286
235,888
Think about the demand of your product before doing a startup: https://nitter.gq/t.co/cKOnqpx8cn
2
1
4
One can comfortably maintain only 150 stable social relationships 'Dunbar's Number' #RelationshipGoals #tbt
2
80% of your Revenue comes from 20% of your Clients 'Pareto Principle' (80/20 Rule) #startup #entrepreneur
3