Gopc

3.4K posts

Gopc banner
Gopc

Gopc

@Gopick19

Cat dad || Expect tech, sports, and hot takes on your feed || CS @IITKanpur || Engineer at GS

Katılım Mayıs 2017
530 Takip Edilen133 Takipçiler
Gopc retweetledi
Yana Boyko
Yana Boyko@_yanaboyko·
Please help spread the word: I’m honestly shocked to see my illustration being used on T-shirts sold at the Monte-Carlo Masters without my permission or a licence. I never expected something like this from such a major tournament. If someone from the organisers sees this, please contact me so we can resolve this properly. @montecarlorolex @atptour
Yana Boyko tweet mediaYana Boyko tweet media
English
360
4.9K
27.9K
1.6M
Gopc
Gopc@Gopick19·
@karpathy isn't this what alphaevolve is supposed to do? Idk how well it works for this use case but what is the difference in ideas?
English
0
0
0
12
Andrej Karpathy
Andrej Karpathy@karpathy·
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.
Andrej Karpathy tweet media
English
962
2.1K
19.5K
3.6M
Tanay Kothari
Tanay Kothari@tankots·
@Gopick19 @WisprFlow Hey, it seems like you downloaded a phishing app by a developer called Butterfly AI. Can you double check and make sure you're downloading one by Wispr AI Inc?
English
2
0
4
704
Tanay Kothari
Tanay Kothari@tankots·
We offered 5 people a Porsche 911 GT3 RS if they could get @WisprFlow to make a mistake It's the fastest and most accurate AI voice dictation app that's 3x more accurate than ChatGPT, Claude, or Siri. Today, we’re finally launching on Android. Download now: play.google.com/store/apps/det… As a part of the launch, we’re giving away 6 months of Wispr Flow Pro for free. Like, retweet and comment ‘Wispr Flow’ to get it. Enjoy. — Written with Wispr Flow
English
4.5K
3K
10.8K
4.3M
Gopc
Gopc@Gopick19·
@tankots @WisprFlow Thanks Tanay! Installing the right one now. Will get back with the results
English
0
0
0
104
Gopc retweetledi
Doctor
Doctor@DipshikhaGhosh·
This is why we need more women in positions of power, more representation in all spheres. Because women are the only true protectors of women. Men can claim to be protectors but all they do is protect their image and protect their brothers who harm women.
Baba Banaras™@RealBababanaras

India : At 10 PM, a foreign woman got lost, alone and terrified after Google Maps failed. With no one around, Rapido rider Sindhu Kumari stopped, reassured her, and safely dropped her to Hotel Coconut, turning fear into relief. Salute to this brave Indian woman rider.

English
145
6.9K
57.2K
744.7K
Gopc
Gopc@Gopick19·
Wrote about AI and how it is changing the nature of work in tech companies An Era of Taste and Craft: Notes on the Post-Mastery Age gopikotana.com/2025/12/17/an-…
English
0
0
0
20
Gopc
Gopc@Gopick19·
What a brave girl!
ANI@ANI

#WATCH | Nanded, Maharashtra | A woman, Anchal, applied vermillion on her head with the blood of her boyfriend, Saksham Tate, who was allegedly killed by her father and brother. Anchal says, "We were together for three years. My family got to know about it. Because he was a Scheduled Caste, my family did not agree to our marriage... My family had told him that if he wanted to marry me, he would have to convert to Hinduism. He was ready to do this also... My family was just waiting for an opportunity to kill him... We had talked in the morning when he was going to the station to drop off his aunt. Even I had no idea this would happen... I got to know about this the next day from the newspaper. No one told me about it... The day Saksham was killed, my brother had taken me to the police station in the morning to get a false case registered against him. I did not agree to file any case. The policemen told my brother that instead of fabricating cases, why don't you actually kill the concerned person before coming to us? My brother took it as a challenge and killed Saksham..."

English
0
0
0
46
Gopc
Gopc@Gopick19·
@Cerebrone That's the record, not the average. Even among the pros
English
1
0
9
1.9K
Gopc retweetledi
Jeet Mashru
Jeet Mashru@mashrujeet·
A Pani Puri vendor from Mumbai's Mulund area has dragged a local politician to court after not providing him a stall space on footpath despite paying ₹3 Lakhs to him. The vendor alleged that the local politician has sold the footpath space to a dosa stall vendor at ₹17,000 per month and broken his deal to buy the space at a total of ₹5 Lakhs. BMC's response - Assistant commissioner says - I have not received any complaints and nor do I have any information about it. Politician's response - You come and meet me, don't publish the story. Let's talk. This is to malign my image. Story by @DiwakarSharmaa in Mumbai Mirror today.
Jeet Mashru tweet media
Diwakar Sharma@DiwakarSharmaa

FOOD VENDOR DRAGS NETA WHO ‘SOLD’ HIM A #FOOTPATH TO COURT Complainant alleges Shinde #Sena leader took Rs 3L on the pretext of transferring the space, but later ‘allotted’ it to someone else. #Mulund #Encroachment #Corruption

English
245
1.5K
5.9K
645.3K