Split View is already supported on @comet. Update it if you haven't yet. https://t.co/I4Rza7zZiC

Use AI to find winning ads for your product I built a Comet prompt for @perplexity_ai that harvests the top Meta + TikTok ads, ranks them and then returns the top 10 performers What you’ll get: - Swipefile with links to winning ads - Public Score ranking (Days active, variant count, view/CTR signals) - Frame-by-frame scripts, captions, CTAs Fill out the template and hit run Follow + Like + RT + comment “ADS” and I’ll DM you the template. MUST BE FOLLOWING so I can DM you.

NEWS: Perplexity has recently overtaken Grok on web traffic, with 168 million visits the last 28 days. https://t.co/znJ2qpgjPH

NEWS: Perplexity has recently overtaken Grok on web traffic, with 168 million visits the last 28 days. https://t.co/znJ2qpgjPH

Pretty cool. @comet is the future. https://t.co/KGIrqeeGtE

🐐🐐🐐 https://t.co/raPJw0oV9F

ICICI bank net banking portal also comes without advertisement while opening from @comet , thanks a lot, earlier so many irritating ads about cards, gold funds, SIP @AravSrinivas https://t.co/l3s1nyr2sW

BREAKING: Cristiano Ronaldo used Perplexity to prepare his Prestige Globe Award speech 🤯 “This afternoon I sat at the table thinking about what I would say in this speech. I said to myself: ‘What is the Prestige Globe Award? Is it a career-ending award?’ I felt a bit nervous and said: ‘It can’t be that.’ Then I did some research on Perplexity—if you don’t know it, look it up later—and it helped me a little.”

Just completed my Coursera course , thank you @AravSrinivas @perplexity_ai @comet https://t.co/QZmnnuLBtl

Telling @perplexity_ai's Comet browser to block the bots in my notifications tab is a great unlock. It performs all actions, going back into the feed, finding another bot, then going to their profile and blocking them. It's not fast, but I can work in another tab while it's running. @AravSrinivas - thanks for making this great product!

I prefer not to speak. https://t.co/0Qbo7vmgST

Wake me up when your comet agent attempts previous year jee advanced questions 🥱 https://t.co/EMKA5R1H9x

On one hand: don't anthropomorphize AI. On the other: LLMs exhibit signs of gambling addiction. The more autonomy they were given, the more risks the LLMs took. They exhibit gambler's fallacy, loss-chasing, illusion of control... A cautionary note for using LLMs for investing. https://t.co/WTnM2T1pAT

This paper shows that you can predict actual purchase intent (90% accuracy) by asking an LLM to impersonate a customer with a demographic profile, giving it a product & having it give its impressions, which another AI rates. No fine-tuning or training & beats classic ML methods. https://t.co/3owtMMZ0m9

+1 more

Big progress on this important benchmark (but still weird artifacts). https://t.co/f9tpCi9fne

Sora 2: "An elaborate Regency romance where everyone is wearing a live duck for a hat (each duck is also wearing a hat). Also a llama plays a flute." https://t.co/7QJnaSIze4

"April 1805, Napoleon is now master of Europe. Oceans are now battlefields. Ducks are now boats" https://t.co/QjwqPLo95V

"An elaborate regency romance where everyone is a duck wearing a tiny human for a hat (each tiny human is also wearing a hat)" https://t.co/n27FB4af9J

Does AI improve or undercut academic scholarship? Using AI increases both the quantity & quality of academic scholarship and reduces inequality: -Researchers using AI published 36% more papers -There is also rise in the journal impact factor of adopters’ publications -GenAI also helps level the playing field in academia. The strongest productivity gains appear among: Early-career researchers + Authors from non-English-speaking countries https://t.co/ifuXAldRTF

AI is apparently already accelerating science. Measuring academic publications of authors: “we find that productivity among GenAI users rose by 15 percent in 2023 relative to non-users and further increased to 36 percent in 2024” and the quality of publications also went up. https://t.co/9DewOj8CkN

Why do you need ai privacy: Reason 132 https://t.co/aAaSZZ64HL

Used my last $35 to get this trading platform https://t.co/c9Bw6RFP1H

We evaluated Gemini 2.5 Deep Think on FrontierMath. There is no API, so we ran it manually. The results: a new record! We also conducted a more holistic evaluation of its math capabilities. 🧵 https://t.co/NOtVQVGHdV

Benchmarking is Broken - Don't Let AI be its Own Judge "In high-stakes human examinations (e.g., SAT, GRE), substantial effort is devoted to ensuring fairness and credibility; why settle for less in evaluating AI, especially given its profound societal impact? This position paper argues that the current laissez-faire approach is unsustainable. We contend that true, sustainable AI advancement demands a paradigm shift: a unified, live, and quality-controlled benchmarking framework robust by construction, not by mere courtesy and goodwill. To this end, we dissect the systemic flaws undermining today's AI evaluation, distill the essential requirements for a new generation of assessments, and introduce PeerBench, a community-governed, proctored evaluation blueprint that embodies this paradigm through sealed execution, item banking with rolling renewal, and delayed transparency. Our goal is to pave the way for evaluations that can restore integrity and deliver genuinely trustworthy measures of AI progress."

Better Together: Leveraging Unpaired Multimodal Data for Stronger Unimodal Models "We introduce UML: Unpaired Multimodal Learner, a modality-agnostic training paradigm in which a single model alternately processes inputs from different modalities while sharing parameters across them. This design exploits the assumption that different modalities are projections of a shared underlying reality, allowing the model to benefit from cross-modal structure without requiring explicit pairs. Theoretically, under linear data-generating assumptions, we show that unpaired auxiliary data can yield representations strictly more informative about the data-generating process than unimodal training. Empirically, we show that using unpaired data from auxiliary modalities -- such as text, audio, or images -- consistently improves downstream performance across diverse unimodal targets such as image and audio."

project page: https://t.co/n9graSSxYg code: https://t.co/bTDN0U98ct abs: https://t.co/9epn0tdavL

+1 more

Agent Learning via Early Experience "training agents from experience data with reinforcement learning remains difficult in many environments, which either lack verifiable rewards (e.g., websites) or require inefficient long-horizon rollouts (e.g., multi-turn tool use)." "We address this limitation with a middle-ground paradigm we call early experience: interaction data generated by the agent’s own actions, where the resulting future states serve as supervision without reward signals. Within this paradigm we study two strategies of using such data: (1) Implicit world modeling, which uses collected states to ground the policy in environment dynamics; and (2) Self-reflection, where the agent learns from its suboptimal actions to improve reasoning and decision-making"

https://t.co/aQsW59Ylt6, combined with learning help from an LLM, is more accessible than ever. Great way to learn how AI and deep learning really work at the foundations. https://t.co/VzJswBMym6

The https://t.co/JWOQGkTeIJ course is how I got started in AI. Few years later, in a full circle moment, I got the opportunity to contribute & help teach a few lectures of the course too. I still think it's one of the best resources for learning deep learning. Highly recommend!

3 minute, REAGENT FREE, CHEAP, virtual stained scans, that have already shown us phenomena like cancer cell resistance. gigabytes of data per cell, piping directly into our AI models. https://t.co/AwnGWSVDUg

Mouse studies often fail to replicate in surprising ways. Let me enumerate some of the ways: 1. Mice handled by male scientists feel less pain. The finding holds true when a female scientist does the experiment but holds a t-shirt, previously worn by a man, near the mouse. The effect fades after 30 minutes. 2. Mice spend less time licking an irritated part of their body when a human is nearby, “even if that ‘person’ is a cardboard cutout of Paris Hilton.” 3. Mice raised at five different animal facilities in Europe, under otherwise identical conditions, had “persistent differences in body weight” and behavior. 4. If two different scientists at the same university carry out the same experiment on mice, their results will be MORE replicable than if the same exact experiment were carried out by the same person at separate universities. 5. Mice that give birth in cages with little toys or knick-knacks produce more pups. Those pups are larger after 21 days. 6. “Mice housed on deep bedding had smaller adrenal, kidney, liver and heart weights as well as larger body and tail lengths compared with groups kept on shallow bedding” after just 12 weeks, according to one study. 7. Animals stored on higher shelves are more stressed and have impaired immune systems, probably because these areas are closer to lights and vibrate more. Mice on the top shelf of a rack receive 20-80x more light than mice housed at the bottom. 8. Mice exposed to even dim light during the night (e.g. an LED on a computer monitor) “had a body mass gain…about 50 percent more than other mice that lived in a standard light-dark cycle.” 9. After just four weeks, mice exposed to a dim light during the night ate more than those in complete darkness. (Mice, like humans, raid the proverbial refrigerator when they can’t sleep.) Many genes linked to inflammation were also activated. 10. Mice kept in cages with wood chip bedding eat about 1.5 grams of their mattress every day. This changes the bacteria in their microbiomes. 11. (Not about replication, but) about 80% of drugs are tested only on male mice. Even though some drugs, notably Ambien, are more potent in females and cause more side effects. 12. About 6% of all mouse genes are regulated in sex-specific ways. The expression level of more than 1,000 genes varies between males and females, and the level of another 600 genes wobble, up and down, during a female’s estrous cycle. 13. Grain-based food usually contains unknown amounts of phytoestrogens, which change the onset of an animal’s puberty. 14. The standard diet for mice, called AIN-93, hasn’t changed in 30 years. But manufacturing of that food HAS changed: Even if you use “the same grain-based diet used in the past by others, its composition will likely differ.” In other words, the same food used in mouse studies today vs. the 1990s is different, even if its name is unchanged. 15. Mice exposed to a regular, 37 Hertz magnetic field spend less time exploring open spaces, and more time sleeping. 16. Mice are kept in rooms between 69 and 79 degrees F. “But the natural comfortable temperature for mice is warmer — between 30 and 32 degrees Celsius (86 to 90 degrees Fahrenheit).” Colder mice experience more stress, their tumors grow faster than mice kept in warm rooms, and “mice genetically modified to develop obesity only gained a lot of weight at warmer temperatures but not at colder temperatures.” All links at source below. These studies make for fun reading materials.

Wow, I just reached a new milestone: 1000 CITATIONS! 🥳 I've been waiting for this one... Onto publishing more research! https://t.co/QxVLmXct3x