asfaan murthyulas

@tunguz Use all thinking patterns that created aha moments in great people who solved impossible problems.

"Solve Riemann Hypothesis. Make no mistakes."

@stats_feed His blood circulated without friction and organs worked without wear and tear.

The inventor of Vaseline was so convinced of its healing powers that he ate a spoonful every day and had himself covered in it when he got sick. He lived to 96.

@andrewwhite01 Generalization of solar system. As above so below thinking.

In the history of the atom, there was a short time that elliptical electron orbits were thought to be the best model (Bohr-Sommerfeld). Elliptic orbits look awesome so everyone kept it around and it's become the default visual (like the emoji - ⚛️). Totally wrong model though

@AllenInstitute May be their function is same. Like gestalt unification.

Neurons don't connect randomly. In this video by our #ElectronMicroscopy team, a blue neuron's axon forms a connection to a far neuron. Along the way, it links to some neighbors while skipping thousands. Connectomics seeks to understand what makes those connections special.

@_YifanGao What about punishment does it travel back to predictable cue.?

New paper!🧠We showed that the Reward Positivity backpropagates from feedback to predictive cues during reinforcement learning — first time this has been demonstrated with noninvasive EEG in humans! Huge thanks to my mentor and my co-authors! Open access: doi.org/10.1111/psyp.7…

@SiniiMayo പ്രത്യേകിച്ച് മതവും രാഷ്ട്രീയ പാർട്ടിയും.

നമുക്ക് ഇഷ്‌ടമല്ലാത്ത ഒരാളെ കുറിച്ച് ഒരാൾ വന്ന് കുറ്റം പറഞ്ഞാൽ നമ്മൾ ഹാപ്പിയായി കേട്ടിരിക്കും അതേ സമയം നമുക്ക് വേണ്ടപ്പെട്ട ഒരാളെ പറ്റിയാണ് പറയുന്നതെങ്കിൽ നമുക്ക് ദേഷ്യം വരും ബേസിക് സൈക്കോളജി അല്ലേ ഇത് ?

@sanjaykumarpv ആന ഇവളെ കണ്ട് പേടിച്ചൊടിക്കാണും

ഉത്തരാഖണ്ഡിലെ ഖാതിമാ റോഡിൽ അദ്ധ്യാപികയായ യുവതി ഓടിച്ച ഹോണ്ട ആക്ടീവ റോഡു കുറുകെ കടക്കുകയായിരുന്ന ആനയെ ഇടിച്ചു. ഇടിയുടെ ശക്തിയിൽ ആനയുടെ കൊമ്പ് ഒടിഞ്ഞു. എനിക്ക് മനസിലാകാത്തത് വഴിയേ പോകുന്ന ആനയെപ്പോലും മര്യാദക്ക് കാണാൻ പറ്റാത്ത ഇവരൊക്കെ റോഡിലൂടെ പോകുന്ന മനുഷ്യരെ എങ്ങനെ കാണും.

@PrinceDavies55 @jon_d_doe Claude Mythos is doing all patch works in cyber security.

@jon_d_doe "Meta laid off 8,000 and shifted 7,000 into AI. You know what that means for cybersecurity? More AI making decisions. Fewer humans catching the mistakes. More attack surface. More risk. Upskill or get left behind."

Meta is laying off 8,000 employees (10% of its global workforce). And re-assigning 7,000 to focus on AI development. In the next few years, it's likely to lay off 50% of its employees. Omo, there is a big threat to a lot of jobs in the future. Big one. End.

@jun_song What is size of your gpu

Frequent question : Are you going to work for a big tech? No. I just left corp for open source. I've received offers from multiple companies, including a frontier lab, but I turned them all down. I will stay independent and dedicate myself to Sovereign AI open source. 💪

@bitcoinmalayali Fine-tuning open source is all you need.

Show me your GPUs.

Trump: The Iran conflict will end soon.

@dilipjain077 Evolved for of free food and free procreation.

Survival instincts of a garlic bread 😂

@rand_longevity List me 10 great qustions

what is the first question you will ask the AGI?

@JaibyGeorge5979 വളം യുദ്ധം കഴിഞ്ഞാൽ എത്തും, പക്ഷെ പറഞ്ഞത് തിരിച്ചെടുക്കാൻ ആവില്ല.

കഷ്ട്ടം ....ഇന്ത്യയുടെ ഒരു ഗതികേട് 😡😡😡

@AutismCapital He is a great guy. How can you add dirty smile. 🤌

You thought YOUR social anxiety was bad? 😂 💀

@manoramanews America is a luxuries country they are proud about. Certainly They will find a way to stay.

'60 ദിവസത്തിനുള്ളില്‍ പുതിയ ജോലി കണ്ടുപിടിച്ചോ, അല്ലെങ്കില്‍ യുഎസ് വിടണം'! ചങ്കിടിച്ച് ഇന്ത്യന്‍ ടെക... Read more at: manoramanews.com/gulf-and-globa… #us #india

@_avichawla Every feature in path to reward become good indicators to maximize.

Karpathy's prediction about RL is coming true now! He called reward functions unreliable and argued that a single reward number is too low-dimensional to teach an agent what "good" means for complex tasks. To solve this, Agents need a knowledge-guided review as a higher-dimensional feedback channel. Every major AI lab trains models with RL today (OpenAI, Anthropic, DeepSeek). And their key bottleneck has always been the reward functions. GRPO by DeepSeek worked well for math and code because the environment gave a binary signal. But for real agent tasks, someone still has to hand-code the scoring function. That takes days and breaks every time the pipeline changes. RULER (implemented in OpenPipe ART, 10k stars) addresses the exact problem Karpathy identified. The reward criteria are defined in plain English, and an LLM evaluates each trajectory against that description to provide feedback for training. I trained a Qwen3 1.4B agent that plays 2048 using GRPO with this exact workflow. In this case, the agent saw the board, picked a direction, and RULER evaluated the outcome, all from this natural language definition. You can see the full implementation on GitHub and try it yourself. Here's the ART Repo: github.com/OpenPipe/ART (don't forget to star it ⭐ ) Just like RLHF replaced manual rankings and GRPO replaced the critic model, natural language rewards are replacing hand-coded scoring functions. RL reward engineering is now prompt engineering. I wrote a full walkthrough covering RL for LLM agents, from RLHF to GRPO to RULER, in the article below.

@SkyNews Its Same as kicking out aged zoo animals to forest.

UK-based bank to replace 'lower-value human capital' with AI trib.al/COxXrSi

@dwarkesh_sp @ericjang11 I think scaling can solve this problem. Leela zero uses transformer architecture. Cnn and rnn have proximity bias but scaled transformer can reach cnn and rnn level performance. Biases can save lots of compute and data( comes with cost of reasoning beyond bias)

.@ericjang11 tried using transformers for his Go bot, but they couldn't beat ResNets. The reason gets at something general about architectures. ResNets are biased towards the local. Nearby things matter more, and a useful pattern in one place is a useful pattern anywhere. Transformers are biased the other way, towards global context, with every position able to attend to every other. Most Go fighting is local, and a useful local pattern learned in one position can be applied anywhere in the board. A ResNet's inductive bias means it gets these insights about Go for free. But a transformer has to pay for them.

@thePartyPartyUS @voooooogel They may contain quality reasoning tokens.

@voooooogel not publishing the unabridged is near criminal

unfortunately openai didn't publish the unsummarized chain of thought, but the summary is 125 pages! the model reaches the crucial idea (which it describes as 'frightening,' i would love to read the unabridged chain of thought here...) on page 39