Ben Cohen

@MillionInt Sol is so so bad like this right now. Feel my brain melting when i read any of its output

RLHF raters definitely got AI hacked by text sounding smart = being smart. In 2026 smell of AI is text that sounds incredibly complicated with not much substance. That too will pass but for now…

I wonder how many days of my life I've spent watching protein bar ads on Instagram

@Techmeme Microsoft should buy @OpenRouter and @appliedcompute

Sources: OpenRouter has discussed a potential sale to a bigger tech company that could value it at billions of dollars, a premium to its $1.3B valuation in May (The Information) (Visit Techmeme dot com for the link and full context!)

Can you guys stop shilling your enterprise AI finetuning slop cos and just make a robot that does my fucking laundry already?

@akothari Compute is not super elastic, probably very hard to estimate exact amount you will need. Could still be making good progress. Also other companies already have consumer products that demand much more inference

My taco bell order was so disturbing that Fable’s classifier blocked it

@jparkjmc Best pizza in the world

when i raise a series A i’m taking my team to chuck e cheese bro i dont care

People finally learning how Reinforcement Learning works from first principles

On today’s episode of Animal Planet, we venture into the depths of Silicon Valley to observe a new ailment plaguing the human race called AI Psychosis

@Varungupta You sound miserable

Buying a football team is pure vanity, with zero additionality to the world. Instead, you found a new university that would live on for millennia, endow 1000 soccer academies to make the US actually good at soccer, build a grander version of the World Trade Center / the Oculus, or, or, or...

@pronounced_kyle I am not absolutely right

New interview question: what important truth do very few LLMs agree with you on?

@fortworthchris great interview, thanks Chris!

Today I sit down with Steven Campisi, co-founder of Hillpointe, one of the largest developers of attainable housing in the country. Most apartment developers design a brand new building for every project. Hillpointe has built the exact same building 800 times - which creates advantages overtime that compound. That repetition plus their material and labor strategy is how Hillpointe builds across the Sun Belt for 30-40% less than everyone else. We get into their direct overseas sourcing business, how they've built their own in-house subcontractor base, how they underwrite new opportunities, the state of the market, and why they run the whole operation as a series of discretionary funds. Timestamps (00:00) Intro (02:58) Steven’s Three-Pronged Competitive Advantage (10:29) Prototype Building Strategy (12:39) Direct-to-Crew Labor and Why Crews Choose Hillpointe (20:02) China Sourcing Operations, Scale & Factory Relationships (25:22) Labor Market Shortages & Specialized vs. Generalist Trades (35:01) Land Acquisition Strategy & Regional Development Teams (41:12) Underwriting Deals (44:09) Achieving Investor Alignment (54:45) Fund Model vs. Traditional JV Equity (1:01:25) Raising Institutional Capital (1:09:48) Debt Structure, Supply Glut, and the Road to a 2027 Rent Rebound (1:16:09) AI and Centralizing Operations With an In-House Contact Center

@JJJ_DPOY Both are better than the sixers process starting 5

bruh this look like 2 separate real starting 5s

@forgebitz spotify's been so buggy lately lol

“4,500 production deploys a day” is spotify adding new songs in git storage or something

Oliver Nickell June 29th, 1993 – June 14th, 2026 Oliver Tree Nickell, beloved son, brother, grandson, nephew, cousin, and friend, passed away in Rio de Janeiro, Brazil on Sunday, June 14, 2026, in a tragic accident at the age of 32. Known as Oliver Tree, he was an artist, storyteller, singer, songwriter, producer, director, editor, actor, filmmaker, entertainer, traveler and creative force; he had just performed in São Paulo as part of his 7-continent world tour. Oliver was born in Santa Cruz, California. From the beginning, Oliver always had the most astonishing and relentless creative drive. He began producing skits, music, and drawings as early as 5 years old. He spent ten years racing BMX, cycle cross, cross country, mountain biking, downhill racing and free-style dirt jumping. Oliver was a DJ in his teenage years and started performing shows under the name "Kryph", where he opened up for big artists in the dance scene. At 17, he signed with R&S Records, under the moniker "Tree", where he put out his first official body of work "Splitting Branches”. During this period, he attended San Francisco State University and later graduated from the California Institute of the Arts (CalArts) in 2017. At CalArts, Oliver created the “Oliver Tree” project, which included producing, directing, acting, touring, film-making, and song-writing. In 2016, “Turbo” was born, a character he developed for the internet, who was loud, colorful, bright and hilarious. That same year, he released his first single on all platforms as Oliver Tree, in collaboration with Whethan titled "When I'm Down". The song became a huge success and went gold. In 2017, Oliver signed with Atlantic Records and shortly after released his first viral hit song “Alien Boy” which was produced with his childhood friend Casey Mattson, as well as frequent collaborator Imad Royal. It was a song that truly encapsulated the Oliver Tree brand. Oliver then went on to release his debut album “Ugly Is Beautiful”, which hit the top 15 of Billboard 200 and reached number 1 on Billboard‘s Top Rock Albums and Alternative Albums charts. Oliver also released a deluxe version of the album which included the song "Life Goes On", a track that was produced with long-time friend and collaborator Getter. This song propelled Oliver into a global audience and was a commercial hit. Followed by the success of Oliver’s song "Miss You”, with Robin Schulz, which became a worldwide hit and was nominated for Best International Song (The BRIT Awards) in 2024. Oliver never stopped reinventing himself. Each album featured a distinct character in the Oliver Tree universe: Turbo for “Ugly is Beautiful” in 2020, Shawney Bravo for “Cowboy Tears" in 2022, Cornelius Cummings for “Alone in A Crowd" in 2023 up until "Love You Madly Hate You Badly” in 2026, where he removed the fake personas and decided to show the world his true self for his final album which was 100% written and produced by Oliver. Oliver had a deep curiosity about the world and wanted to experience every country and culture. Over the past two years, Oliver was a global citizen, visiting over 100 countries: eating local cuisine; connecting with locals of all kinds; participating in ceremonies; giving concerts; and recording music on his laptop. He went to the Middle East, sailed in Antarctica and became an honorary Maasai tribal chief in Tanzania. Oliver was more than a musician; he was a true artist in every sense of the word, seeing the world as a stage for his performance art. He pushed boundaries in every form of media and life. Oliver was on a mission to bring the world together through art and inspiring other artists to create. Being a forward thinker, Oliver created a non-profit entitled "Dr. Oliver Tree's Extremely Epic Art Grant for Baby Geniuses”. He wanted all of his money to go into funding other aspiring artists in all facets: music, film, installation and performance art projects. Oliver cared so deeply about spreading art, joy, laughter and love into the world. He blended his characters seamlessly with his authentic self, which left the audience often wondering what is real or a prank. He found beauty in the ordinary and would constantly preach his "Ugly Is Beautiful" mantra on and off stage. He wanted everyone to feel like they had a place to be themselves. Under all the absurdist humor and outlandish stunts, he wanted to create an inclusive environment where everyone belonged and could realize their true potential. Oliver was a creative force of nature, a true “Alien Boy” among human beings. His imagination was boundless, his laugh was contagious and his creativity and ideas were prolific. Oliver is survived by his parents, Jesse & Christine Nickell of Santa Cruz, brother Jessup (Zoe), Grandmothers Ann Begin & Lorraine Nickell, Aunts Cynthia Begin, Mia Begin (Bill Schroeder), Uncle Dan Begin (Melissa), Aunts Sheree Kouffeld (Dave Christopher) and Jan Lamascus (Marvin), many beloved cousins, Beth, Breanna, Kaitlin, Kimmy, Maggie, Meadow and Nicolas; and numerous friends worldwide. He considered his team and colleagues his family: Dan Awad, Paul Donatelli; his bandmates: Casey Mattson, Amir Oosman, Jake Jamieson, Jmsey; his creative collaborators: Ethan Snoreck aka Whethan; Ryan Farber; Steve Zilberman; Jacob Dennis; Sebastian Hackett; his love Fiona Chernavskaya, as well as many others who touched his project and helped him execute his vision, who are too innumerable to be named. We’ll leave you with a quote which Oliver said at almost every show he performed at, his mantra to the world: “No matter how strange you think you look, no matter how ugly you feel, you are beautiful.” A celebration of his life / memorial service will be held at the UCSC Quarry Amphitheater on July 25th. Due to limited space for family and friends, this event will also be streamed live. Donations: The family requests donations be made to “Dr. Oliver Tree's Extremely Epic Art Grant for Baby Geniuses”. Oliver’s vision is to support and encourage young artists to follow their dreams. See more at olivertreefoundation.com

@dwarkesh_sp Check out pedagogical RL for opsd

What does the next training paradigm look like? 0:00:00 – The big research bet the labs are making 0:02:12 – Grindability is just as important as verifiability 0:06:10 – Will RLVR alone generalize? 0:08:41 – Getting the learning back to the weights 0:15:22 – Dreaming 0:17:23 – What 2027 looks like Also on YouTube, pod feed, and Substack.

@urboldforthat @signulll im not arguing the end state. I'm saying the current branding of the whole AI industry is not resonating with people outside of SV

@blc_16 @signulll It will soon become with scale. It’s simple unit economics. Driver has to be paid per hour. AI does not. Losing jobs sucks and policy must adapt. But is it really the endgame for humanity to be driving metal toys around? We can do better things with our time, no?

what a fucking tragedy for nyc. the same dumb resistance happened during the uber saga & now you’re watching the same shit play out again with autonomous cars. the reality is that humans should not be operating metal death machines forever. esp in a city where taxis have spent decades gaming routes, ignoring pickups, driving like maniacs, & treating the meter like a side quest. uber fixed some of that, then slowly became expensive as hell, degraded, & unreliable in its own stupid way. collective decision making of humans is often so so damn stupid.

@urboldforthat @signulll its not cheaper lol. Theyre like double the price of ubers in SF and put many people out of work

@blc_16 @signulll -Significantly cheaper -Safer especially for women (Uber receives an SA complaint every 8 minutes…) -More reliable (no drivers canceling rides/gaming the system, ride estimates are exact) -Faster -Electric In what world are Waymos bad?

@___Harald___ you're not allowed to have takes this reasonable on X sir

I submitted a write-up about intelligence and sample efficiency to Dwarkesh's blog post contest a while ago. Posting it here since the ideas are topical again: Sample efficiency, credit assignment, and self-improvement. That is what a machine system needs to be good at to replace humans. State-of-the-art machine learning is somewhere between bad and terrible at all 3. I will try to convince you that progress will be difficult in the RL regime, and progress so far isn't as great as advertised. Measuring progress in AI is difficult. There is a lack of clear definitions, and an abundance of loud voices driving hype for the next fundraising cycle. This causes a microscopic focus on small advancements in benchmarks, instead of a look at the bigger picture. AI progress has not had a visible impact on any real-world trendlines such as GDP, life expectancy, or farm yields. All the growth trajectories look steady and undisturbed. Modern ML algorithms are getting better at learning from small datasets than ever before. But compared to humans, LLM training runs are incredibly inefficient. A human can read 5 tokens per second, that means it takes 10 years to get to 1B tokens. You might argue visual or auditory channels can allow faster learning, but this seems unlikely given a deafblind person like Helen Keller can become an accomplished author. Karpathy's nanochat was trained on 11B tokens and can't hold a real conversation. Frontier models do pass a Turing test, but they are trained on well over 10T tokens. That's 10,000 times as many tokens than a human can read in a lifetime. When your data is abundant and cheap, this sample inefficiency isn't a problem. LLMs were incredibly useful, even before reasoning and agentic breakthroughs. But a lot of the magic these LLMs provided was just access to incredible datasets. I would have been thrilled to have all the books ever written indexed and searchable on my home computer in 2015, and this would have been technically feasible. But due to copyright law, we needed to wait for the LLM training loophole to actually get this as a product. This brings us to the RL regime. RL is needed for agents to discover new behaviours, and it's what brought us all the coolest achievements in AI. Chess and Go were solved by some form of RL years ago, and now RL is critical in making the agentic LLMs as good as they are. All these successes come with caveats. Chess and Go have cheap, low-dimensional environments that make exploration trivial. For LLM post-training, many of the strategies can barely even be considered RL. RL with Verifiable Rewards is the most exciting strategy used, but even that usually requires guidance. What all these successes have in common is fast feedback loops and cheap environments. Modern ML is simply not capable of dealing with sparse samples and delayed rewards. They will get better, but it will take a lot of innovations to be competitive with humans. These issues are more obvious when looking at the state of robotics. The most impressive robots are Waymo and FSD, which still rely mostly on classical supervised methods. Humanoid dancing robots use RL to be robust to noise when executing dance moves, but those moves are usually precomputed trajectories collected from humans with motion capture. We are far from having the robots come up with the moves themselves. Self-improvement is barely even worth mentioning. The best LLM agents today are not capable of running a datacenter. Even given time to prepare, an autonomously run datacenter would likely fall apart within days or weeks, let alone improve itself. I suspect it wouldn't take more than a few social engineered emails to bring down the entire operation. The progress in AI has been incredible, and I am happier for it. But we must stay grounded. DNA is the best optimizer to have ever existed, it has been unbeaten for billions of years. I hope that we can some day build its successor, but it will take more than a few tricks.