Y Combinator
Everything Y personally uses, recommends, or has created — plus things they don't recommend — sourced from their own show and appearances on other podcasts.
Top picks
Ranked by repeated personal use, excluding sponsorships.
Personal use in 1 episode
"Codex 5.5 is now good enough. You can oneshot most simple internal software dashboards to a pretty high level of quality. I tried it over the weekend on a bunch of our stuff. It's just unreal."
▶ 10:55Warned against
"It's better than Siri and I'm like, why isn't Siri terrible? Siri so bad. Even just the voice input on it is so bad."
"it chose Whisper V1, which is a model from several years ago. Like, the API is practically deprecated... It was like literally takes an hour to process an hour of video"
Recent episodes
All →Showing 124 of 139 products
Clear filtersMentioned in 1 episode
"Tri Dao did this GPU tiling idea called flash attention, which is a huge innovation."
Mentioned in 1 episode
"Even better, if it's hot in DRAM or if it's in some Redis cache, some SSD somewhere."
Mentioned in 1 episode
"DeepSeek reports that it took about 2.8 million H800 hours to train DeepSeek V3, and they do about 43,000 H800 hours per day."
Mentioned in 7 episodes
"If you have like my mom, who is constantly starting brand new chats with ChatGPT that have nothing to do with the past... there is no KV cache."
"What was the capability of those compared to what you can see now? Those code-as-policies papers came out in 2022, right around when ChatGPT came out."
"Maybe these days they're overrelying on ChatGPT and Claude. GPT doesn't like consumer. Maybe if you had an idea and said, 'Where should I go to get funding?' it would probably tell you enterprise."
"In 2020 with GPT-3 we had sort of the emergence of in-context learning."
"Real science breakthroughs will be augmented by AGI, and now's the time that's going to happen."
"when you ask a question on ChatGPT or Claude or pretty much all of the major LLMs"
"when I asked Compliance to let us use ChatGPT, I didn't even get a response"
Mentioned in 1 episode
"Saranovas is your AI operating partner. It reads every record your company produces, finds the operational inefficiencies nobody's tracking, and warns you while problems are still small."
Mentioned in 1 episode
"First, looks like we got Ego. Agents are machines. We make them people. Ego is an applied AI research lab building personal agents indistinguishable from humans."
Mentioned in 1 episode
"Okay. We got Sentient OS. Your Mac, sentient. An on-device LLM understands your entire life then proactively offers to get your work done through computer use."
Mentioned in 2 episodes
"And so, yeah, we had Cloudflare being able to stop DDoS's, but everything breaks when you apply infinity to the number."
"The winning companies will look more like Cloudflare than Rathon."
Mentioned in 1 episode
"I actually think that a great idea for a startup is like an Apple App Store of benchmarks where there's curation, right? They don't just let anyone just throw an app onto the App Store. They go through a process."
Mentioned in 1 episode
"There was this paper in 2023 called Thousands of AI Authors on the Future of AI... and none of that is in the next 10 years. And this is the paper that most people are quoting."
Mentioned in 1 episode
"There's a lot of systems where we just saw this with ICLR, where 60,000 submissions were submitted in 1 year."
Mentioned in 1 episode
"If I had a really smart always-on AI system that's looking at traffic and looking at my Datadog, looking at whatever is coming through..."
Mentioned in 1 episode
"There's a famous physics paper called More Is Different, and it's kind of like the premise... And I think in this world where we had 7 billion people and now we have 700 trillion agents... more is just so different."
Mentioned in 1 episode
"We already have Dan Fu committed from Together AI to talk about inference."
Mentioned in 1 episode
"We've all seen the demos online of it controlling Blender and making great 3D images."
Mentioned in 1 episode
"Inspired by this paper also from Philip Isola that he titles the Platonic Representation Hypothesis, where Platonic is a reference to Plato's Cave. The point he's making is that there's extensive evidence to show that language models and various representations of different data actually learn these distance mappings between similar things under different training policies that are overlapping."
Mentioned in 1 episode
"Prime Agent and Nvidia, which both recently just came out — with some harness, this thing that doesn't deserve any research, just some wrapper and some scaffolding, we can get to 95 and AVO from Nvidia got to 100%."
Mentioned in 1 episode
"Whatever actual inference engine you need to run it. So this could be Ollama, Llama CPP, VLM, SG Lang."
Mentioned in 1 episode
"We decided to propose Open Jarvis. We wanted to explore just how much of this we could run on device completely for free while preserving security, privacy and quality."
Mentioned in 1 episode
"I don't know if you guys read Situational Awareness when it came out in like 2024. But that was kind of an era when test time compute was just starting to become a thing and we're starting to give agents tools for the first time and there's sort of this intuition that there's a little bit more intelligence in the models than we're using in a lot of cases."
13 recommended · 2 sponsored · 2 warned against · 142 episodes