InstructGPT, every mention
28 scenes across 11 shows · ← back to InstructGPT
tap a year for its mentions
Latent Space 13
No Priors 4
TBPN 3
the Y Combinator Startup Podcast 2
Lenny's Podcast 2
20VC 2
the Neon Show 2
We Live to Build 13 more shows
every year every show Latent Space 13 No Priors 4 TBPN 3 the Y Combinator Startup Podcast 2 Lenny's Podcast 2 the Neon Show 2 20VC 2 We Live to Build 1 
A Product Market Fit Show 1 
Sourcery 1 
the MAD Podcast 1
Verbatim, from the transcripts: passages where InstructGPT comes up on Latent Space, No Priors, TBPN, the Y Combinator Startup Podcast, Lenny's Podcast
The Man Training GPT, Gemini & Claude Reveals What's Coming Next | Vijay Krishnan, Turing
- ▶ 5:54 Vijay Krishnan So I'd say, uh, um, I mean, openly I wrote this instructor, this famous instruct GPT paper in early, um, uh, that, uh, showed that, uh, um, uh, relatively small, uh, volumes of, uh, high quality, um, human generated data or high quality…
Mercor Head of Product on Revenue Concentration from Frontier Labs
- ▶ 15:33 unnamed speaker So the, uh, shape of data has changed a lot since it started, um, with InstructGPT for Gen AI, from supervised fine tuning to preference ranking to all these environment type projects.
- ▶ 42:34 unnamed speaker Uh, SFT was really hard to get set up when instruction, uh, when InstructGPT first came out.
Why Coding is the Fastest Path to AGI | Turing CEO Jonathan Siddharth
- ▶ 9:56 Jonathan Siddharth In supervised fine-tuning, there's this paper from OpenAI called InstructGPT, which talks about the recipe for how to do this, where, uh, you're teaching the model to be an assistant.
Everything Gets Rebuilt: The New AI Agent Stack | Harrison Chase, LangChain
- ▶ 36:11 Harrison Chase I think like the DaVinci models and stuff like that.
[State of RL/Reasoning] IMO/IOI Gold, OpenAI o3/GPT-5, and Cursor Composer — Ashvin Nair, Cursor
- ▶ 17:18 unnamed speaker The idea was, you, you started with codecs, someone else was doing instroft GPT, then we launched GPT, four, four O, I guess O-one.
How Josh Kushner Became The Most Powerful VC in New York
- ▶ 14:36 unnamed speaker That's, that, that needs to be, like, corrected, because GPT-III came out in 2020, so I think they're either talking about GPT-III.V, DaVinci-III, which was really popular, or it's just a preview of ChatGPT, because investors were getting…
Inside The $2.2B AI Research Accelerator | Turing · Sourcery with Molly O'Shea
- ▶ 19:59 Jonathan Siddharth The next step is post-training, and OpenAI came up with this paper called Instruct GPT, which lays out this process in more detail.
He quit his job, went all-in on AI agents—then grew to 100K users & a $30M Series A in a year. | ... · PMF Show
- ▶ 9:08 Soham Ganatra Uh, been playing around with, like, like, like, GPT-II, then, like, sort of, such a GPT, DaVinci, like, all of those things coming, uh, and, like, at some point, I thought, like, I felt enough confidence that, okay, this thing is improving…
Weekly Recap: CEO Affair at Coldplay Concert, Elon Musk's AI Girlfriend, Cognition Acquires Windsurf
- ▶ 2:19:10 Martin Shkreli So when, when DaVinci came out and, uh, I was in jail when GPT, GPT two came out and I romped through the jail phone, uh, and it was, uh, pretty humorous.
DWARKESH PATEL on the Biggest Problem of AI
- ▶ 4:04 unnamed speaker See you in GPT-III, GPT-III.V, DaVinci, ChatGPT, and it starts feeling like, okay, we passed the Turing test, we need to really have this conversation about AI, and then, and then PDOOM and AGI becomes like the main discourse for like a…
Information Theory for Language Models: Jack Morris
- ▶ 3:06 Jack Morris Like around when I guess GPT three, one hundred and seventy five billion had been released, but not instruct GPT.
- ▶ 1:11:12 Jack Morris The InstructGPT techniques, you just need the data. 2 times in the scene
Cursor CEO: Going Beyond Code, Superintelligent AI Agents, And Why Taste Still Matters · Y Combinator
- ▶ 23:36 Michael Truell And, you know, we still remember all the launches of, you know, InstructGPT, which made GPT-III a little bit better.
⚡️Open Questions in Agentic RL — Will Brown (Prime Intellect)
- ▶ 2:46 Will Brown Um, and so we've kind of moved beyond the single-term RL eject world of, like, InstructGPT, ejectGPT sort of things, as well as the
The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using | Michael Truell
- ▶ 1:01:21 Michael Truell You know, InstructGPT doesn't exist. 2 times in the scene
Training Llama 2, 3 & 4: The Path to Open Source AGI — with Thomas Scialom of Meta AI
- ▶ 5:11 Thomas Scialom I actually was working on Galactica Instruct, which basically you could connect it, we had a partner with Overleaf, the Google Doc of, like, scientists, where you can write papers, and it's in, you write there in LaTeX, you have to do a… 2 times in the scene
No Priors Ep. 65 | With Scale AI CEO Alexandr Wang
- ▶ 6:34 Alexandr Wang yeah, they integrate this, uh, all of this technology into GPT-III with, uh, there was a paper InstructGPT, which is kind of the precursor to ChatGPT that we worked with them on.
High Agency Pydantic over VC Backed Frameworks — with Jason Liu of Instructor
The Origin and Future of RLHF: the secret ingredient for ChatGPT - with Nathan Lambert
- ▶ 25:15 Nathan Lambert It has this three-step process that we'll go to, into more when I kind of go into the main concepts, but it's like the first time you see this diagram that they reuse with InstructGPT, they reuse with ChatGPT, and the types of examples… 2 times in the scene
- ▶ 43:35 Nathan Lambert I threw in the, um, slide 28, it's like the InstructGPT metadata.
- ▶ 58:11 Nathan Lambert I think InstructGPT does something where they, like, try to get the RL model to match these, the instruction tuning model, or the instruction tuning dataset, because they were really happy with that dataset to constrain the distribution.
Beating GPT-4 with Open Source Models - with Michael Royzen of Phind
- ▶ 14:51 Michael Royzen Um, this is before InstructGPT.
550 Million People, One Language, Zero Access to Venture Capital
- ▶ 32:23 Ricardo Michel Reyes It's good because it uses reinforcement learning from human feedback, which was discovered in the StrapGPT paper two or three years ago.
No Priors Ep. 15 | With Kelvin Guu, Staff Research Scientist, Google Brain
- ▶ 10:57 Kelvin Guu But because that worked out, things like InstructGPT came out next, followed by things like ChatGPT, and this sort of adaptation is
No Priors Ep. 3 | With Stability AI’s Emad Mostaque
- ▶ 10:32 Emad Mostaque The next stage is the reinforcement learning, the instruct models, flan palm or instruct GPT or others that requires very specified annotation.
- ▶ 29:39 Emad Mostaque They announced InstructGPT, their 1.3 billion parameter version outperformed one hundred and seventy five billion parameter GPT-III.
The REAL potential of generative AI · Y Combinator
- ▶ 5:54 Raza Habib In the InstructGPT paper that OpenAI released, they compared, you know, a one or two billion parameter model with instruction tuning and RHF to the full