Yogi coder

@neural_avb How would you rate it for reasoning based tasks ? Like long sets of text and images to go through and reply in structured way

Holy Deepseek Pro 51M total tokens in a day and spent just $0.92 (most of it was cached input)

@drishtadyumn Oh fuck off Pure and absolute BS To anyone reading, putting random number on charts does not mean they are true. x.com/i/status/20489…

Kids these days don't know, but before Narendra Modi took over the country, the concept of heatwave was non-existent! You could easily survive summers without AC but Modi and his industrialist buddies conspired to cut all trees so they could sell more ACs Horrible!

@kushal_mehra Can we talk about universities and the shit they allow to take place ? Not just India but Canada as well

Guys if you work in an MNC and are being victimised share your story. There are enough media outlets and podcasters that have a conscience and will speak up for you irrespective of their political leaning. Please speak up. And remember sources are protected.

@yifever 29% haven't watched Veritasium video

which one do you think is a bigger technological feat

@neural_avb I'll keep an eye for its repo

Damn I am vibecoding something that's gonna help me so much to produce tutorial content faster opentui + ipykernel + vim motions + ffmpeg A jupyter notebook directly in my terminal that will auto run code and create an mp4 from... No need for me to physically screen record anymore after this!

@MilksandMatcha With robotics and to have hardware + software platform for students to learn from first principles. This access won't be for me but for her. (and then me pushing her to augment her teachings with codex/AI)

@MilksandMatcha Building multiple projects while also dealing with health issues. One of the projects is about AI assisted robotics classes with a close relative. She teaches coding to young children (often under privileged) to make a foundation level. This summer vacation she's planning big 1/n

Giving away 5 Codex Pro plans Each person will get 3 months of free Codex Pro (highest tier). Winners will be selected from comments in 48 hours, comment below why you want it.

@neural_avb I've been deferring to learn fine tuning for a while now . You have videos on those ?

I've been working on a finetuning tutorial series. Staring with part 1 on Continued Pre-training (CPT). I am taking an 135M param SLM and training it on ~150 new Arxiv papers. Been playing with chunking strategies, dataset sizes, mixing datasets, full ft vs lora etc. Also testing multiple eval metrics (ROGUE, Perplexity, BLEU, BERTScore) etc. A lot of cool learnings so far, I need to sit down and consolidate everything that should go in a video.

@seergioo_gil Kinda doing the same though am building a module that selects which ones to animate and highest impact. Agentic graph runner of sorts Not using I2V though still all automated. (I know audio sucks in this one, was using a free version)

animated 1 of each 5 scenes of this vid for $0.30c scenes animations are now integrated into the timeline so you don’t need to create a whole new template for it (this is low-model btw, and one shoted)

@ThePollLady @desimojito @thehawkeyex and now you too ? Who's paying for it ?

Saving a few dollars on illegal IPTV can cost thousands in fraud recovery. Identity theft isn’t instant it lingers. Legal streaming avoids that risk entirely. #IdentityProtection #Streaming

@ShivAroor What did CD/DVDs say when you'd burn them in Nero ?

I miss CD-ROM drives. That whining sound as the tray slid out and back in. Remember “burning CDs” and those big stacks of Moser Baer CD-Rs, where most would be faulty? Sigh. Good times. 💿♥️

@levelsio Claude's got attitude too with lying

@karpathy Ah yes, one step closer to be done with data scientists / ML eng But regardless, New play thing found for this week

oh yeah i should have linked autoresearch probably github.com/karpathy/autor… (you don't "use it" directly, it's just a recipe/idea - give it to your agent and apply to what you care about.) and the tweet about it that went mini-viral over the weekend with more context x.com/karpathy/statu…

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.

@noahmorris Not a lot with video styling but trying to bridge gap between "templated" vs human edited videos with generative fill Very nascent atm using remotion studio ,generating and/or filling in assets within a scene But object tracking would be cool ! Though u can animate assets here

Just had a major tech breakthrough on vidrush. Main goal for this month is to introduce generative editing meaning it simulates micro decision making the way a real editor would including per video styling, pacing, J & L cuts, A roll, object tracking, verification of subjects and more. The term “Vidrush channel” will soon disappear as you won’t be able to tell the difference. The main focus after this is releasing our talking head feature as I do very much believe that long term most of YouTube faceless will move into a mature stage of having actual hosts on the channels I hope to introduce this mid to end of April 😄

@noahmorris New hermozi style title: "I survived brutal warzone and built a hundred million dollars faceless empire"

Welp first time having to hide underground from missles lol

@PhedEU Nice project for the weekend

I am now producing 20 minute 3D documentaries in minutes.. for $5 per video welcome to the dead internet era

Hi team @VidrushAI @noahmorris @sonerabay Guess I solved something which looks promising.Would be happy to get feedback 🙂. Have a few more versions in store as well ~2mins/$1.2 max,excluding audio as I used free stuff Cost can be further reduced by half 100% automated

Dear West, Read and reform. 1. Epstein files may be shocking. But think deeper - child exploitation has historic roots in western culture, unlike Vedic culture. 2. Romeo in Romeo Juliet was not even a teenager. Every other fairy tale has a witch that wants to kidnap or kill a child because she is more beautiful. 3. Beauty and Beast, Snow White, etc etc - they become part of what you teach kids. 4. Children falling down from a hill becomes Jack and Jill nursery rhyme. 5. Humpty Dumpty, which a childish face, shattered into pieces after falling down. Kids are supposed to enjoy this. 6. The entire "free-market" colonialism which made West prosperous was rooted in child-exploitation. Adam Smith wanted children above 8 years of age to be labor. Special machines were designed to make children work. 7. On contrary, Vedic or Sanatan or Hindu culture equated children with God. Be it Baal Gopal. Or worship of little girls as Devi in Navaratri. 8. Before Bharat came in contact with invaders and got contaminated, children safety was the root of civilization. 9. It was a declared rule that - Children, Women, Old, Diseased, Mentally ill and Scholars will never be harmed. 10. A daughter in a village was considered daughter of entire village. Romances within village is a western contamination accentuated by foreign-inspired Bollywood. Else, it is unthinkable. 11. It is time that west rejects the regressive exploitative roots and embrace Vedic Sanatan source. This is source of their language and whatever is good in their society. 12. Yes, Vedas are the parent. Bharat is the original light. And distortions are the result of deviations. 13. Come back to source. Leave the path of so-called missionary conversions. Evangelism has worst track record of human rights violation and exploitation that is no way less than what Epstein Files exposes. Welcome to Satya Shiva Sundar Sattvik Sanatan Vedic Dharma. Welcome to reason. Welcome to eradication of hypocrisy. Sanjeev Newar

Something to look at and boost @thehawkeyex

@noahmorris @onlinedopamine 🤣

@onlinedopamine 🫠

whenever you see anyone on your feed talking about arcads, higgsfield, rork, and the likes, just know that this is the reason why