Greg Kamradt's banner

Greg Kamradt

@GregKamradt • 50,977 subscribers

President @arcprize, builder/engineer

Shorts

my new vibe code setup: 1 orchestrator agent which controls 85 sub-agents working in parallel each sub-agent spawns from my stream of consciousness and tests from the main orchestrator Here's how it works:

my new vibe code setup: 1 orchestrator agent which controls 85 sub-agents working in parallel each sub-agent spawns from my stream of consciousness and tests from the main orchestrator Here's how it works:

316,172 views

315x1 Bench - New 1RM Been chasing this one for 4-5 months Now time to move onto the next goal

315x1 Bench - New 1RM Been chasing this one for 4-5 months Now time to move onto the next goal

46,514 views

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

> Last year, I was constrained by tokens. I fixed that by joining OpenAI > Then I was constrained by CPU, now I feel my constraint is actually *attention* .Peter Steinberger 🦞 came to agents/pizza/wine to demo his agentic engineering workflow

> Last year, I was constrained by tokens. I fixed that by joining OpenAI > Then I was constrained by CPU, now I feel my constraint is actually attention .Peter Steinberger 🦞 came to agents/pizza/wine to demo his agentic engineering workflow

203,478 views • 2 months ago

How to summarize a book without sending 100% of your tokens to a LLM I tested out a fun method on "Into Thin Air" (~140K tokens) The results were surprising good Definitely enough to refine and keep working Here's a speed run though process:

How to summarize a book without sending 100% of your tokens to a LLM I tested out a fun method on "Into Thin Air" (~140K tokens) The results were surprising good Definitely enough to refine and keep working Here's a speed run though process:

227,592 views • 3 years ago

The mindset shift that changed the way I looked at the world: “I’m going to build my own tools” I've automated more daily workflow in the past 12 months than I have my entire career “You can build your own things that other people can use. Once you learn that, you'll never be the same again.” - Steve Jobs "Coding is dead" - Said no one serious The secret wasn’t running away from code No! The opposite. I dove head first into new tools (Cursor, v0, LangSmith, OpenAI, Anthropic, etc.) As the CEO of Cursor puts it, we are entering the golden age of “delight concentration” We can now be creative instead of technical. The friction to building your own tools is literally just a prompt away (Anthropic Artifacts). For custom tools, the barrier to entry is mind-boggling low. Tools like Cursor, v0, shadcn, LangChain and Vercel allow you to build, deploy and share what matters to YOU faster than ever It’s no longer skill level that's holding you back, it’s just your desire to take a step forward

The mindset shift that changed the way I looked at the world: “I’m going to build my own tools” I've automated more daily workflow in the past 12 months than I have my entire career “You can build your own things that other people can use. Once you learn that, you'll never be the same again.” - Steve Jobs "Coding is dead" - Said no one serious The secret wasn’t running away from code No! The opposite. I dove head first into new tools (Cursor, v0, LangSmith, OpenAI, Anthropic, etc.) As the CEO of Cursor puts it, we are entering the golden age of “delight concentration” We can now be creative instead of technical. The friction to building your own tools is literally just a prompt away (Anthropic Artifacts). For custom tools, the barrier to entry is mind-boggling low. Tools like Cursor, v0, shadcn, LangChain and Vercel allow you to build, deploy and share what matters to YOU faster than ever It’s no longer skill level that's holding you back, it’s just your desire to take a step forward

117,317 views • 1 year ago

zen is watching 42 agents work across your screen watching them land is my new asmr yes, I made flight radar for my AI (full screen, sound on)

zen is watching 42 agents work across your screen watching them land is my new asmr yes, I made flight radar for my AI (full screen, sound on)

57,616 views • 1 year ago

visualizing text splitting & chunking strategies ChunkViz .com I thought I remembered a tool to visualize text chunking, but I couldn't find it, so I built one I didn't realize it would be so visually pleasing to tinker with 4 different splitters featured

visualizing text splitting & chunking strategies ChunkViz .com I thought I remembered a tool to visualize text chunking, but I couldn't find it, so I built one I didn't realize it would be so visually pleasing to tinker with 4 different splitters featured

116,228 views • 2 years ago

Well, it happened. I accidentally became obsessed I found a business that built a Q&A app using Slack messages as a knowledge base Years of tribal knowledge locked behind keyword search LLMs make it useful Way bigger opportunity than I thought Here's 40 companies building:

Well, it happened. I accidentally became obsessed I found a business that built a Q&A app using Slack messages as a knowledge base Years of tribal knowledge locked behind keyword search LLMs make it useful Way bigger opportunity than I thought Here's 40 companies building:

81,396 views • 2 years ago

Low key one of my favorite presentations of the year Sam Whitmore sharing her "AI Memory" journey building New Computer * How her AI Memory/RAG approach changed 2023 > 2025 * Why the perfect memory architecture doesn't exists * 4 types of memory (queried in parallel)

Low key one of my favorite presentations of the year Sam Whitmore sharing her "AI Memory" journey building New Computer * How her AI Memory/RAG approach changed 2023 > 2025 * Why the perfect memory architecture doesn't exists * 4 types of memory (queried in parallel)

34,115 views • 1 year ago

How good is GPT-4-Vision at extracting text from images? I wanted to find the limit - but I found weirdness instead Most surprising: GPT-4V performance varies depending on the *structure* of text it sees Let me explain A set of images with progressively more text was presented to GPT-4-Vision. GPT-4V was asked what text it saw in the image. The response from the model was compared against the image’s original text and scored for similarity. The model was tested with 4 types of text: essay, random words, random tokens, and random characters. Findings: * Performance degrades - Yes, the models are good at basic OCR, but as you get more text and words then performance drops (this is expected) * Type of context matters - You should expect different recall on your texts based on your context types * Hallucination Errors - I thought that the model would make errors of omission (it wouldn’t return all the words). But instead the model mostly made hallucination errors - it replaced words with made up words. * Evals Matter - This test in isolation doesn’t mean that your data will have the same results, but it should motivate you to create eval tests for your data and anticipate errors which are hard to spot Notes: * Next step would be to add additional image types like tables or PDFs * GPT-4V would routinely get stuck in repeat-token-loops when trying to extract random tokens * GPT-4V would refuse to answer most random character images

How good is GPT-4-Vision at extracting text from images? I wanted to find the limit - but I found weirdness instead Most surprising: GPT-4V performance varies depending on the structure of text it sees Let me explain A set of images with progressively more text was presented to GPT-4-Vision. GPT-4V was asked what text it saw in the image. The response from the model was compared against the image’s original text and scored for similarity. The model was tested with 4 types of text: essay, random words, random tokens, and random characters. Findings: * Performance degrades - Yes, the models are good at basic OCR, but as you get more text and words then performance drops (this is expected) * Type of context matters - You should expect different recall on your texts based on your context types * Hallucination Errors - I thought that the model would make errors of omission (it wouldn’t return all the words). But instead the model mostly made hallucination errors - it replaced words with made up words. * Evals Matter - This test in isolation doesn’t mean that your data will have the same results, but it should motivate you to create eval tests for your data and anticipate errors which are hard to spot Notes: * Next step would be to add additional image types like tables or PDFs * GPT-4V would routinely get stuck in repeat-token-loops when trying to extract random tokens * GPT-4V would refuse to answer most random character images

49,111 views • 2 years ago

Relistening to sales calls is so March 2023 I built Thimble to help teams extract data from sales calls, powered by Here’s the journey so far:

Relistening to sales calls is so March 2023 I built Thimble to help teams extract data from sales calls, powered by Here’s the journey so far:

44,583 views • 3 years ago

Just tried ChatGPT agent on ARC Prize ARC-AGI-3 > Told it to play a game > Couldn't figure out what to do > Agent did a web search, "how to beat X arc-agi-3 game" > It didn't find answers > I told it to try clicking red/blue blocks > It clicked them, noticed something happened, kept clicking > Nudged more > Couldn't figure it out > Searched again Then I cut it off btw agent is a very cool tool

Just tried ChatGPT agent on ARC Prize ARC-AGI-3 > Told it to play a game > Couldn't figure out what to do > Agent did a web search, "how to beat X arc-agi-3 game" > It didn't find answers > I told it to try clicking red/blue blocks > It clicked them, noticed something happened, kept clicking > Nudged more > Couldn't figure it out > Searched again Then I cut it off btw agent is a very cool tool

15,386 views • 1 year ago

No more content to load