Harsha Bommana

@ziikarmali Gaming!!

looking for my next idea inspiration. in my "fafo" stage. any suggestions?

@27upon2 Can you build an agent to do this tho?

Understanding failure modes by looking at agent trajectories is so fun

We're still figuring out how to build a great research lab. This is a huge step in the right direction! Thinking about AI research with these mental models has completely changed the way I approach and value research contributions. It's so different from engineering!

I recently sat down with Front Page by the @Analyticsindiam to unpack @ylecun's new LeWorldModel We discussed: • Representation collapse • The exact difference between the LLMs and World Models • The implications of LeWorldModel youtube.com/watch?v=sy8WtC… @AmitRajaNaik

We are very glad to announce @AnthropicAI as the first sponsor for our Conference for AI Scientists (CAISc) 2026, co-organised by Lossfunk along with @bitspilaniindia. Top three spotlight papers at the conference will be awarded $2,000 each worth of model credits. This is very exciting for us because scientists across the world have been doing with amazing things with Claude and we believe these awards are a great chance for motivated teams to meaningfully multiply their research output. Check out how senior scientists have been using Claude 🧵

Exciting times ahead! 🚀

@navbenny @RuntimeBRT Yooo LFG!!! 🚀

Happy to be covered by @RuntimeBRT! This is the era of the Intelligence Arbitrage. No longer will the markets be dominated by speed and information asymmetries, in the future, the markets will reward the most intelligent models. But unlike general purpose LLMs, this will need specialized models that understand the language of the markets: price. That is what we are building. Over the last year, our breakthrough foundational models have been managing our fully autonomous hedge fund delivering industry leading alpha. There is no human in the loop and we don't know why it does what it does, but it is just incredible to watch! It's been a long journey but I'm excited to keep pushing the frontier and see what lies ahead :)

📢 [New Preprint] VLMs often struggle to play video games. Is the bottleneck reasoning or perception? We gave three strong VLMs object coordinates while they played Atari, VizDoom, and AI2-THOR. Perfect coordinates helped every model. Self-extracted coordinates only helped one VLM, indicating perception to be the bottleneck. Accepted at LM Reasoning Workshop, AAAI 2026

@raj_chandra3 @Sameeran_B Bro that’s Sourav

@harshablast Feel realistic like @Sameeran_B 🤩

We made these posters for a friend on his birthday for party decor Top: Last year, we had to finetune a flux model with LoRA for a few hours on his face images Bottom: This year, we just prompted Gemini AI progress is crazy

Brilliant way to probe frontier LLMs' actual reasoning abilities. Even with esoteric languages, given the documentation & examples, models should be able to answer questions, right? But they can't! Check out this work by @inceptmyth for more details!

Day by day, I'm losing my faith in the concept of backpropagation

@inceptmyth I tried using claude code to analyze my experiment logs (around 50 plots/images for each run). It does a great job at giving me an overview of the data, but fails to give me fine-grained insights that I would have missed as a human. Might be better with a custom agentic workflow

AI agents are smart and they can answer questions (great at data analysis/ visualisation, they can come up with insights based on logs of the experiments, they can find common patterns quickly) but still they suck at asking the right questions. I was running an experiment on curriculum learning for "something that will be coming very soon" and it does plane old data analysis without coming up "what do these analysis really mean for us, how can we integrate and build on top of this and what should we be even looking for in these dense logs of rich data." If as a researcher you don't ask the right questions, you won't even get the sniff of the answers you were looking for. Not sure how this can be automated in a seamless "auto research" way because not all research is pure optimization in parameter space.

Kickstarted Batch 7 of @lossfunk residency! 🧵 with their profiles and goals for the next 6 weeks

Giving a talk at @lossfunk tmr with @vitransformer! This will be about hierarchies in visual representation learning. We’ll be doing a deep dive on two pivotal papers in this field. Do register at luma.com/m9fenrve if you’re interested in attending!

@SarahLevinger You've nailed the disconnect! At PerzAI (perzai.app), we understand that the customer journey is far from linear. Our AI-driven tools analyze real buyer behaviors to create content that resonates at every stage of their actual journey.

How marketers think people buy: 1. Buyer sees ad. 2. Buyer clicks ad. 3. Buyer purchases. Versus how people *actually* buy:

@loshminft Definitely! The key to impactful content is uniqueness and personal touch, which is often lost in today's template-driven strategies. We're building Perz AI (perzai.app) to ensure that every piece of content is tailored and engaging, turning generic into genuine.

Content Creation has become repetitive and oversaturated, is it something that we can fix? I believe that last few months of Content Creation from all the sides have been consistently repetitive, don't think that's the case? Let me elaborate: How many times have you been scrolling on X just to see a same tweet, same template, same project being shilled over and over again? I believe for the Web3 as whole to grow we need to change that fact especially because in most case scenarios people won't even take a minute to read the thread anyways. The simple reason why they don't even bother reading is because they've already seen the same tweet at least 5 times that same day. For a project to have a good marketing it doesn't necessarily have to be focused only on getting Content Creators/Influencers to make a tweet for them. If your product is good and you have a good backing of your project as a whole and have a vision towards that you're looking to build around, people that are investing through conviction plays will definitely come around and check your project out. What we also lack is unique content creation, many people out there don't even try to share the actual knowledge they've learnt last years and help others out with the information they have. Instead what they do is pretty much copy paste already done tweet by somebody else, because they see that tweet as a good way to "farm" engagement. In conclusion, for us to have a little change on how we approach X Content Creation we have to stop doing the same things and focus on actual unique content.

@jakesnowfall Couldn't agree more on the pitfalls of copying content. That's why PerzAI (perzai.app) focuses on crafting personalized and unique content that resonates with each brand’s audience, ensuring authenticity at scale.

Avoid these common content creation mistakes: 1) Not Practicing: you need to put in the reps 2) Over-Consuming: create 80%, Consume 20% of the time 3) Copying Everyone: We don't multiple people posting the same stuff Be original and authentic

@Chris_Martine2 At Perz AI (perzai.app) creating personalized content using data-driven insights is our mission. Please check it out and let us know what you think! We would love to hear your feedback and insights.

In 2024, social media platforms are doubling down on AI and machine learning to create personalized experiences for users. From custom feeds to tailored content, the future of social media is all about delivering what you want, when you want it. #AI #SocialMediaInnovation #DigitalMarketing #DigitalMarketingTips