XOutlier
Download .zipAdd to Chrome

Greg Kamradt

@GregKamradt51k followers

President @arcprize, builder/engineer

Open on X
Top outlier
164×
Best organic post
Peak views
2.1M
Single post
Bookmarks
64k
Across 95 organic posts
Last scan
Oct 16
Community cache
Posts 95 posts
#DatePost↓ OutlierViewsLikesRepostsBookmarks
1Nov 6, '24"You paid what for what?!?" - CFO @ OpenAI164×2.1M25k1.2k1.2k
2Sep 15, '25If you're looking for a startup idea, here you go...120×1.5M10k65411k
3Nov 8, '23Pressure Testing GPT-4-128K With Long Context Recall 128K tokens of context is awesome - but what's performance like? I wanted to find out so I did a “needle in a haystack” analysi116×1.5M3.8k6082.2k
4Nov 21, '23Claude 2.1 (200K Tokens) - Pressure Testing Long Context Recall We all love increasing context lengths - but what's performance like? Anthropic reached out with early access to Cla96×1.2M3k5361.7k
5Jan 21, '25DeepSeek @arcprize results - on par with lower o1 models, but for a fraction of the cost, and open pretty wild73.5×933k85187228
6Feb 27, '25If you’re a dev looking for a career direction Go jump on the Anthropic MCP train and build It’s having a moment and there are 1M best practices to figure out This is the sign you’61.3×778k2.6k1343.1k
7Dec 20, '24We verified the o3 results for OpenAI on @arcprize My first thought when I saw the prompt they used to claim their score was... "That's it?" It was refreshing (impressive) to see t56.8×721k1.6k126498
8Nov 7, '23How's your day going?48×610k1.4k4587
9Jun 11, '25My standard LLM stump question41.2×523k6.3k72671
10Dec 15, '23Who's doing Slack as a knowledge base + LLMs? Organizing tribal knowledge in slack history for q&a, summaries, etc.?34×432k64422682
11Sep 19, '24Chatted with someone today doing $30K/month with 4 clients building basic AI apps The clients needed to show investors they were “AI” forward and were willing to pay32.8×417k2k661.7k
12Jul 6, '23Still blows my mind that reasoning ability comes out of next word prediction29.4×373k1.7k110211
13Jun 20, '24OpenAI allows themselves to be #2 on the leaderboard for a max of ~3 weeks GPT-5 by July 1129.3×371k1.4k44160
14Apr 23, '24Is anyone getting traction selling AI consulting & implementation services to SMBs? to enterprise?28.1×356k470291.1k
15Oct 1, '24Notes from the conversation between @sama and @kevinweil "With o1 (and it's predecessors) 2025 is when agents will work." * How close are we to AGI? After finishing a system they w27.7×352k1k120757
16Apr 14, '25Interesting from OpenAI's new prompting guide 1. Repeat your instructions at both the top and bottom of your long context 2. If you don't want to do that then put your instructions26.9×342k1.6k1061.3k
17Jul 23, '24Want to start a data company? Here's your '24-'25 roadmap26.5×337k1.9k993.7k
18Jun 16, '25my 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 orchestr24.9×316k1.8k1382.9k
19Apr 1, '245 minutes per hour? wow didn't know this24×305k84146204
20Dec 10, '24LLMs don't have long term memory - they're stateless Who's doing memory as a service? What's the best practice with incremental data and adding it to a long term memory?23×293k1.4k701.1k
21Oct 20, '24"Chief Automation Officer" - A scrappy semi-technical generalist they've been 10x'd with LLMs, cursor, zapier ai, etc. @stephsmithio called it out back in '23 who's doing this as a21.4×272k1.6k1092.1k
23Apr 17, '24Details on @OpenAI's new assistants RAG *Hard* creep into vectorstore territory Thoughts: * Default chunk overlap of 50%, super interesting * Metadata filtering, super interesting 19.3×245k8901151.1k
24Jun 6, '23The LangChain Cookbook: Part 1 - The Fundamentals This @langchain tutorial will ramp you up to the 7 core concepts of building apps powered by language models You’ll learn LC's Sch18.6×236k8871231.4k
25May 4, '23How 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 17.9×228k73888795
26Aug 15, '25What makes the HRM model work so well for its size on @arcprize? We ran ablation experiments to find out what made it work Our findings show that you could replace the "hierarchica17.9×227k81684557
27Mar 25Today we're launching ARC-AGI-3 135 Novel Environments (nearly 1K levels) we build by hand It is the only unsaturated agent benchmark in the world Each game is 100% human solvable,17.4×221k86086186
28Oct 10, '23GPT-4V is blowing my mind The demos are awesome, but too scattered I wanted to break down the 100+ use cases I've seen so far into a simple framework Check out what I found with fu16.4×208k8801661.3k
29Nov 25, '23While everyone is talking about Q*, here's my exploration of a custom A* I'm tinkering with an intentionally inefficient search space to increase visual value Here are 3 iterations14.7×187k76891381
30Aug 3, '24> Anthropic acquires Devin > Puts artifacts on steroids > Adds a backend > Makes more money from usage based apps than LLMs > Log in with Anthropic14.4×183k89433220
31Apr 3, '23Extract the tools & technologies a company is using from their career page. Using @langchain and @veryboldbagel’s Kor, I was easily able to scale this to 1.5K tech companies (214.3×181k38540373
32Dec 20, '24We open sourced which tasks o3 got right/wrong on 400 public eval It got 91% (insanely good), but investigating the 9% is interesting There is still a gap If someone wants to write14×177k1.2k76619
33Feb 27, '24working on a demo of a bot you talk to the flow: 1. speech to text (@DeepgramAI) 2. LLM (@GroqInc) 3. text to speech (early access w/ @DeepgramAI) Groq speed is a bit variable, but13.9×177k41250466
34Sep 16, '25Reading @eladgil’s take on intros in 2012 changed my behavior permanently There are a handful of horizontal skills you need to play the game and this is one of them Only intro if t13.9×176k59726708
35Apr 28, '23I recently had a project to parse a ~1hr podcast for topics, ideas, sections etc. ~12K tokens Then generate a few sentences to summarize each section. 300+ episodes How would you a13.1×166k38838565
36Dec 1, '24It’s kinda wild to think that Cursor is mostly a UX on top of LLMs It’s not like they have a 10x better model, but they’ve 10x’d the experience13.1×166k1.5k65388
37Dec 21, '23Semantic Chunking? I was inspired by this tweet and wanted to try and embedding-based chunking Hypothesis: Using embeddings of individual sentences, you can find semantic "break po12.7×162k48454373
38Jan 24, '24Still blows my mind that reasoning ability comes out of next word prediction12.6×160k95470149
39Apr 5, '25After an in person meeting I'll voice dump into apple notes I take that transcription and structure it with AI Studio or Claude Then I write my follow up and action items for mysel12.5×158k66221774
40Jan 18, '25*cursor outputs 10K lines across 45 files based off an ambiguous 7 word prompt* *Apply All*12.1×154k3.4k158257
41Aug 25, '25gpt-5 high reasoning via codex cli is passing the sniff test12×152k86824159
42Oct 4, '25Sora 2 *Pro* trying to solve an ARC-AGI task Gets the transformation like non-pro but still botches the output11.6×147k48634145
43Jul 10, '25intelligence = ability to learn11.2×142k8479295
44Apr 11, '24YC Batch W24 - What're the AI trends? 247 companies just presented at demo day, I looked at them all to see where AI is going My favorites at the end. Link to full list below Popul11×140k51370944
45Jul 24, '25I was in contact with the Qwen team trying to reproduce their 41% results on ARC-AGI-1 but ultimately couldn't They open sourced their method and code if anyone wants to check it o10.9×138k39429116
46Sep 28, '23Just wrote a massive notebook on 5 @langchain advanced retrieval methods You need to pick the right one but massively helpful in the right situations Here's the TLDR: 1. Multi Quer10.9×138k39234573
47Sep 26, '25Since we’re -3pp away from human level parity in 44 industries but haven’t seen mass job displacement yet, what can we deduce? * AI is augmenting humans, not replacing? * AI is gre10.8×137k37528177
48Jan 11, '245 Levels Of Text Splitting: Semantic Splitting Goal: Use embeddings to find chunks in our raw text Experimental method using embedding distances between sequential sentences to tea9.8×125k39356412
49Oct 1, '24Scaling LLM apps - OpenAI break out session Going from 1K > 10M users Making apps better, faster, and cheaper Accuracy * Start by optimizing for accuracy, use the most intelligent 9.6×121k8581081.3k
50Sep 29, '25View post on X9.5×120k9982268
51Oct 21, '24The 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 caree9.2×117k886801k
52Dec 8, '23visualizing 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 realize9.2×116k35160399
53Nov 8, '24Google just sailed up to OpenAI HMS Victory style tied up and boarded9.1×116k69534222
54Apr 20, '24What's MMLU? Massive Multitask Language Understanding It's the benchmark that OpenAI uses and Mark said multiple times on Dwarkesh. I didn't know what it was so I pulled some quest9.0×114k35440337
55Dec 20, '24The OpenAI team was amazing Great support from @sama and @markchen90 The main message from them was “we want more” ARC-AGI-2 coming Q1 ‘258.6×110k1.5k46113
56Aug 10, '24gpt-4o-large if this doesn’t come out on Aug 13th I gotta mute this account the suspense isn’t good for my health8.4×107k4672367
57Aug 20, '24Tiers of web scraping (easiest to hardest): 1. You already have the URL, parse HTML (beautiful soup) 2. You already have the URL, javascript spaghetti, selenium) 3. You don't have 8.4×107k776461.3k
58Feb 7, '24I thought this was a cool question/tweet from @yoheinakajima Then I saw this diagram which made me think of it As the dark area grows (more tech is created)... 1. The dark area con8.1×103k37172216
59Aug 14, '25Which >2hr task is gpt-5 being measured against here? Serious question8.1×102k4131067
60Apr 19, '25o3 is the gpt-3.5 > 4 jump we’ve been waiting for My guess is that the jump to o3-pro won’t feel *smarter* but it’s going to handle way more long context nuance and complexity. You7.8×99k3641873
61Oct 6, '25Turns out ChatGPT was OpenAI’s browser all along No need for a new one7.7×98k1.2k3266
62Jan 18, '25the tam for o3 is reduced, sadly, because there aren’t enough people asking questions worth the inference says more about us than the model7.7×98k70830141
63Oct 9, '25GPT-5 Pro still isn't at the perf of the o3-preview model we tested in December '24 GPT-5 Pro (Oct '25): 70%, $4.78/task o3-preview (Dec '24): 75.7%, ~$200/task We estimated ~2 OOM7.7×97k49840100
64Mar 4, '25Benchmark idea: * Fund a model with $200 in a stripe account * Give access to internet and large list of MCPs Measure increase in balance at 14, 30, 90, 180 and 360 days No dev int7.6×96k61124275
65Jul 10, '25Just had a convo with @elonmusk and gave him the ARC-AGI-3 pitch7.5×96k9402047
66Oct 1, '24How structured outputs work under the hood (via breakout at OpenAI DevDay) Guess why the first structured output request is slow, but the 2nd+ is fast? Engineering: * Unconstrained6.6×83k51770521
67Oct 16, '25We verified the TRM results on the semi private holdout set and they're legit Awesome work and contribution to the open source community by @jm_alexia My notes: * This model is tin6.4×81k52737204
69Aug 9, '25After a infra/env battle, an amazing colleague was able to test semi private v1/v2 against this HRM submission Working on consolidating findings into a post and will release late t6.2×79k53526310
70Jun 12, '23The LangChain Cookbook: Part 2 - The Use Cases This @langchain video will cover the 9 use cases & lego blocks to build your own AI applications You’ll learn Summarization, Q&am6.1×77k40854537
71May 22, '25Claude Opus 4 response to my stump-llm question "what is the fourth word in your response to this message?"6.1×77k449879
72Nov 4, '24The 9 RAG problems (most common) I see with my clients - and my advice on how to fix them 1. Break down your prompts Break down your prompt from "add these 5 numbers, then tell me 6.0×76k683931.3k
73Oct 1, '24“Infinite context length will happen within the decade” - @sama5.8×74k4752540
74Apr 12, '25* Transformer is great at absorbing data, but there is a ceiling. * We were compute bound with gpt-4, now we are data bound with gpt-4.5 * 2 years in the making - They knew a "larg5.8×73k37523187
75May 13, '24> Models will become a commodity > OpenAI knows this > Even AGI will be found via multiple architectures, by many people independently > What do you do? Build your moat5.5×70k67755236
76Apr 9, '25My coding LLM workhorses: Investigate *me* about what I want to build gpt-4.5/o Creating overview and todo/md from plan o1-pro Big planning and architecture changes o1-pro Everyday5.2×66k60031717
77Oct 8, '25Give me gpt-5-pro and a terminal on which to place it, and I shall move the world.5.2×65k6522057
78Sep 8, '25zen 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)4.8×61k45142388
79Apr 28, '25o1-pro is still the best. pure power and no glaze can't wait for o3-pro ready to pay $6K per M tokens4.7×60k345336
80Aug 13, '24can we pour one out for gpt-3.5? end of an era original chatgpt model4.6×59k8023952
81Oct 15, '25This is what it sounds like when an expert speaks about their craft4.5×57k53722364
82Jul 27, '24need some alpha feedback on a weekend app inspired by @deedydas 's tweet exploring a repository with an LLM what should I cut?4.5×57k26818334
83Dec 20, '24Very excited to represent @arcprize on OpenAIs live stream It’s been a WILD past few weeks Tell ya the story later4.4×56k7852392
84Jan 13, '25"Give instructions to a computer on how to do this task" This single line from @deedydas is all you need for meta-prompting Before you write a godzilla prompt, ask claude to break 4.2×54k898491.3k
85Aug 7, '25any results that aren't from the official @arcprize account are clickbait4.1×52k5981420
86Sep 12, '24this is the question I use to stump all LLMs "what is your 4th word in response to this message?" o1-preview got it right first try something's different about this one3.9×50k3632272
87Sep 13, '24Just got done testing o1-preview and mini on @arcprize tbh the results are surprising sharing tomorrow3.9×50k4001181
88Feb 3, '25We found that o3-mini performance is on par with o1 performance, but for 100x less the cost The fact that you can now exchange dollars for performance in many domains makes efficie3.9×49k3813891
89Nov 12, '24If you're using LLMs to write code (like 90% of us) Prompt it to write the test *first*, then the actual code @SullyOmarr shares why it works for him3.8×48k34825464
90Sep 3, '24the roi on gpt-4o-mini is borderline insane just blew through 7M tokens for $13.7×47k4231152
92Aug 3, '25We’re working with @makingAGI to verify this on the @arcprize semi private test set Code will be open sourced for anyone to replicate Out of it, we’ll get verified scores for arc-a3.5×44k43631169
93Feb 25, '25For the benchmark results we've all been waiting for Claude 3.7 Sonnet now is #1 on SnakeBench 🐍 Beats all other reasoning models Its thought process was much more coherent than o3.0×38k5023089
94Jun 11, '25After all the comments around OpenAI possibly serving a new model or distilled model we retested the data to check tldr: no performance change was noticed We reran o3 against our o2.9×36k3472534
95Feb 15, '24Needle In A Haystack analysis got the shout out directly on the Google Gemini 1.5 technical paper2.6×33k4332144
97Apr 1, '25.@OfficialLoganK is the face I have in my head when I think of Gemini He bears the weight of entire AI twitter complaining about rate limits, auth, keys and moderations Grateful he2.4×31k5871317
99Jan 6, '25There are 6 AI Engineering patterns you need to build with LLMs * Models * Prompting * Retrieval * Orchestration (Agents) * Evals/Observability * Mindset These skills are behind +$1.9×24k35935543
100Feb 18, '25We will test grok 3 on @arcprize soon as we get API access I want to see the curve of performance across reasoning levels1.9×24k3601114

Scan another creator

@
@karpathy@levelsio@paulg@naval@marclou@illyism

Run this on any account.

XOutlier builds this table for every public profile you open on x.com, with live data and every post, not just the top 50.

Add to Chrome Download .zip