GPT

“So early days, because these frontier models were very, very strong. So you end up using multiple frontier models, right?... Like Gemini, the Claude, the Cursor, the GPT. But the grounding of that. Will happen with our proprietary data. That data doesn't go an…”

“GPT has really, like, gotten up on Gemini in terms of image generation.”

“We here in this demo that we gave at hot chips we're showing GPT OSS running at over 4000 400 TPS, which is just mind blowing.”

“Most models, I think, are derivative of either, either the you know, like, anthropic series or GPT series. And so, like, You know, those two kind of have two distinct personalities, but if you go use, you know, any of the, like, open source tools or things lik…”

“Where Kimi K-Stri is not as expensive as Claude or GPT Sol, but it is a lot more expensive than JLN-F.”

“They are behind in chips. But their approach was at the next level is open source models where they're producing some extraordinary models. Not as good as GPT or Anthropic or Google's Gemini, but very good.”

“This is a thing that has been tried internally within Google, and it led to Google missing GPT.”

“There's a one question we got to back up and really explore, which is, is AI a new platform? And I would argue that it is not yet a new platform. It is an important technology. We have a new kind of portal to that in GPT or whatever chat we use, AI chat, but i…”

“In fact, like, you know, GPT found a proof to an unsolved Erdos problem, and our competitor Harmonic, you know, Aristotle you know, verified it.”

“We've seen some of the best results with using either Opus from Anthropic or GPT for OpenAI for the orchestrator.”

“So one of the core design principles which we had in the company is every model development has to be a tailwind for us, not headwind. If tomorrow Google launches a new Gemini model or OpenAI launches a new GPT model or like, you know, something like deep secr…”

“Use, use the good models. One, because they're really hardened against prompt injection and some security risks out of the box. And two, you're just going to get a better experience. So I use a lot of Opus four, six, Sonnet four, six, and GPT five, four.”

“A colleague of mine, a peer on, on our product leadership team, Train to GPT on a bunch of transcripts from past product reviews that are all publicly available. And we expect that our PMs are running their PRDs or their pitches through that to say, what's Rac…”

“So we trained the model. Just on our own internal content and then like licensed databases of let's say plant facts or weather or something like that. And so if you ask it a question or send it a picture, it'll give you the answer that the closest answer you c…”

“Most bolt on players are not doing it. They're simply using a GPT or anthropic model. And they're basically just adding a thin layer. You have to rebuild the entire experience end to end.”

“Yeah, I mean, they passed GPT in the app store.”

“And so if I were non-technical, I would want to build this out in just a basic GPT first. And if I was a developer and someone was asking me to come on as a co-founder, I would say, have you built this in a GPT first? And if not, let that be the first thing th…”

“And so then I tried this across models, and I saw consistently across Grok, and GPT, and Gemini, that you were seeing this phenomena where models will, like, self-correct themselves quite a bit, and these were, like, non-thinking models. They were, like, the s…”

“What this suggests about the GPT training data is that the self-reflection data has now actually become pretty much core to the training of all frontier models, because we're seeing that happen in non-instruct models across the board.”

“I think whether GPT, 5.2 codex max or Opus 4.5 is going to do better. It's probably doesn't matter. It's just a matter of like, which one's going to have all the information about what's in the lab and how much will it cost? How long will it take?”

“And at the end of the day, it was just a GPT with the Texas speech model.”

“And what we found was that We were able to add, like, maybe 10, 15% to the battle cards that, that initially was not that clear to us.”

“I think we had about seven million impressions on launch day, and on our website, I think, two years ago, it was about five million, and leads were coming in so fast that I had to build a GPT automation that would qualify leads automatically, because we'd lite…”

“What is interesting with generative AI is that we are being reconditioned to tolerate much longer delays. So if you use something like GPT or you use something like Claude or your favorite chatbot, oftentimes it's just sitting there thinking for seconds and se…”

“What I have personally found, and to be clear, I haven't tried to build a full on application with Gemini three yet, but I think that Claude's earlier models, like I would always go back to them. Like when I was trying to use codex and GPT five before 5.1 came…”

“So I've created Lessons Lessons for Kids. It's on Spotify. I've created episodes like The Man in the Arena, Theodore Roosevelt, Conqueror of Worlds, Alexander the Great, The Man Who Questioned Everything, Socrates, and I basically have built, like, a little pr…”

“And it's like, but it's like, comparing them, like, these models now can, like, actually, like, figure out exactly what you want, and more than 10% of Etsy's traffic is straight from GPT.”

“10% of the world's population is a GPT weekly active user.”

“I believe we're going to a webless world. I believe GPT is going to be like a quarter to a third of the traffic and nine months from now, right?”

“We built this GPT that's trained on SOPs literally loom transcripts, meeting notes, our policies on AI usage, HR docs, any like admin really related question. And it also knows everyone on the team and their role.”

“Basically the model then becomes don't ask questions. You haven't asked GPT first.”

“Where there, there's going to start being GPT's just for this. Like if it's like a brainstorming GPT, right, and it's optimized for that versus like long form copy or short form copy or conversation or coding or, right, there's all these uses people are using …”

“Yeah, I mean, both of them are, the two of them are better at this benchmark than any of the models that we've seen before this week.”

“So we use AWS and Claude and Gemini and GPT and Mistral kind of interchangeably. The biggest thing there has been, how do we build everything in such a way where we can hot swap the models whenever we want, and also build it in such a way that the models becom…”

“There's a company in our portfolio that is doing really well. I was talking to GPT about like, how much follow on capital do you think this company should like, what is the potential outcome here? Like I was doing all of this math with GPT just to try to like,…”

“The only actual real thin GPT wrappers are the apps from these companies like chat GPT is a thin GPT wrapper. Claude is a thin GPT wrapper. Like Claude, the website, Claude, the app, is a thin GPT wrapper. It's an input box and an output box, and a little spri…”

“When I came in, I sort of said, let's just delete all of it, and let's build on GPT, and let's, like, build a version of it that, at the end of the day, lawyers are benefiting from.”

“Cause I think GPT is the, you know, undisputed winner of like the consumer chatbot wars so far.”

“Now I built, like, a super easy GPT that we call, like, the whiz spell checker that is used not by marketing. It's used by every single person in the organization. What it does is it makes sure that it's, like, according to our brand and tone of voice.”

“Historically, what we have seen from our previous initiatives is that, okay, maybe the best GPT or best cloud is the top model, but there might be very small gap with the model just below it. And then it becomes a cost performance trade off so that users can k…”

“With GPT, it's actually neutral, like, it really, it's really good, right, I write down the rules of my chat group, and it's 40,000 people and it's, it doesn't ban anymore, it just mutes people for, like, a day or 10 minutes”

“Maybe even if we tested this with LALAMA architecture, like, there's sort of like a GPT++ architecture, like, I would guess that can store better data just because the kind of numerical flow is a little bit better, the nonlinearities are maybe, like, A little …”

“I, like, I'm telling you, GPT generates better enterprise case studies faster than the typical marketing associate does at a company in, like, one millionth effort.”

“Playing with just the frontier models, Claude, GPT, Gemini you know, it just makes a lot of sense, in my view, and becoming a, like, top one percent user, there's probably a few things that you can do that will be as high leverage as that.”

“It's okay if you're a rapper on GPT, because you want to establish customer value first. If you have established customer, if you are getting customers, everything else will follow, right?”

“On our benchmarks, it performs about 20% worse.”

“Yeah, but still around 40% as opposed to 85% on this particular test set.”

“Anything where it just seems simple and like, you can do one off and you don't need to bring a ton of context into it. You can typically use like sonnet or GPT for that. Anything where I feel like if I was going to try to implement it myself and I would need t…”

“The first order effect here is that GPT means that every person in the world has, like, a median competent tutor or sidekick. This is obviously amazing for education, and if you can't recognize that fact, I'm not sure you have such good judgment.”

“It basically is just content with a different index into it.”

“Yeah, the information search function already, we moved over to like using some GPT based stuff.”

“So, like, GPT doesn't have very good air recovery ability. And so, because of this, it will go into loops and do the same thing over and over and over again, whereas Claude does not do this.”

“A lot of people will make GPT wrappers for things like Cal AI. We have so many clones right now, and most of them have done nothing to make the app more accurate. They just put a basic prompt in and have GPT estimate the calories, whatever that means from the …”

“GPT is all generative based on predicting the next token and patterns and then getting those results to check that they're correct. So I think a lot of it is you had to have a lot of data that was factually correct and Fed into probably the model and the train…”

“And at the time, unsupervised learning was just not really working. So he noticed this one really interesting property, which is there was a neuron that was flipping positive or negative with sentiment. And yeah, that led to the GPT series.”

“That paper was actually cited by OpenAI, and they're still doing research papers. Like, we were told by some of the folks that worked on the first version of GPTs, like one and two, that that actually had influenced them to work more in NLP.”

“My friend Claude, Tells me that it's actually seven percent of iPhone users are parents, and I trust Claude more than I trust GPT.”

“We're not buying the IP by poaching someone from like a lab who tells us how they train GPT. Never done this. Will not do it.”

“There's a whole model that's been trained in a different way. So there's probably a whole lot of tasks it's probably better at than you know, GPT and Gemini and Claude.”

“People say that these things are just wrappers on GPT in some cases, and you're like, well, SaaS is just a wrapper in a SQL database, and that worked out just fine for lots of people.”