Sam Hogan ๐บ๐ธ
@0xSamHogan
Followers
15K
Following
10K
Media
340
Statuses
3K
The inference .net grants program is live!. Apply and get $10k for AI projects that need LLM inference from top models:. - R1 and V3 from @deepseek_ai.- Llama 70b and 405b from @AIatMeta.- High rate-limits, batch support, and more. If you're building open source, read belowย ๐
9
8
78
btw, vercels email unsubscribe form is unauthed. you can unsub anyone
105
91
6K
Apple buys OpenAI and Sam Altman becomes Appleโs CEO in 5 years. Sam leads Apple, the largest company in the world, for the next 25 years. Apple + OpenAI effectively reinvent consumer technology as know it, and the world is forever changed. Sam goes down in history forever.
751
169
5K
The founder of a YC-backed AI customer support Discord bot quoted me $16,800/yr for 1,000 questions per month. >99% markup on the underlying tokens. Letโs see how fast Claude 3.7 + Cursor can build it from scratch while Iโm stuck at home with the flu. GitHub link incoming
100
120
4K
How to get a girlfriend in SF:. 1. Sell to Meta for $100M.2. Sell to Google for $100M.3. $10M on $100M pre product Founder's Fund seed round.4. Buy bitcoin in 2013.5. Buy Ethereum in 2016.6. Move to NYC for a year and bring one back. those are your only options. choose wisely.
133
68
4K
Just tested OpenAI's new Assistant's API. This is now all the code you need to create a custom ChatGPT trained on an entire website. Less than 30 lines ๐คฏ
85
353
3K
LLM chat interfaces need to support "branching" so it's possible to create a new thread with the *current* chat context to ask tertiary questions without polluting the context of future questions.
214
139
3K
@RhysSullivan once you make the decision not to skip, it's unlikely you will change your mind. absolute trash content.
11
1
3K
6 months ago it looked like AI / LLMs were going to bring a much needed revival to the venture startup ecosystem after a few tough years. With companies like Jasper starting to slow down, itโs looking like this may not be the case. Right now there are 2 clear winners, a.
131
419
2K
Introducing Autodelve ๐คโก๏ธ. YC-backed AI customer support tools quoted us nearly $20,000/yr for a simple Discord Q/A bot. So built our own with @cursor_ai in 2 hours and made it open source. Fork the repo, index your docs site, and get answers in Discord in under 5 minutes
45
77
1K
โจ I spent the last 48 hours making GPT-4 read the entire @solana validator codebase and write documentation, so @aeyakovenko doesn't have to. Introducing โ an AI-powered chatbot trained on nothing but code that can answer deep
87
241
1K
Prediction: Microsoft buys cursor for $10B-20B sometime in 2025.
91
34
1K
Letโs see what the smart folks on hacker news had to say about the original windsurf launch
OpenAI Reaches Agreement to Buy Startup Windsurf for $3 Billion ~ Bloomberg. Thoughts??
52
51
1K
Just figured out how to bypass OpenAI's new 20 file limit for Assistants. This one trick from Computer Science will let you train your Assistants on 10 GB of text data ๐คฏ
37
144
1K
POV you're about to hear an HBS grad explain AI agents
today marks a special moment - I'm beyond excited to announce wordware's $30m seed round led by @sparkcapital (followed by @felicis & @ycombinator). for decades, coding was the only way to create in our digital world. but AI is changing everything - it thinks in beautiful shades
24
14
960
SF founders execute 7 days a week. We do not go to art events. Our city is on fire. We do not strategize. Strategy is inaction. We relentlessly march forward, day by day, minute by minute.
after half a year as a NYC founder I've figured out the ideal weekly schedule:.sun: strategize.mon-fri: execute.sat: go to art & tech events in bushwick.
118
31
819
AI-powered B2B SaaS apps are in trouble ๐ตโ๐ซ. Instead of paying almost $20,000 I made an AI-powered support bot in less than 3 hours using only @cursor_ai and Claude 3.7. Fully open-source. Clone the repo and set up your own in less than 5 minutes. This is the future of software.
The founder of a YC-backed AI customer support Discord bot quoted me $16,800/yr for 1,000 questions per month. >99% markup on the underlying tokens. Letโs see how fast Claude 3.7 + Cursor can build it from scratch while Iโm stuck at home with the flu. GitHub link incoming
32
67
814
damn girl you really have non-technical co founder energy.
40
34
764
I spent the last two weeks meeting with 50 teams building AI apps the single most consistent thing I heard was โwe stopped using langchain and other agent frameworks and built our own internally and it works way better and is easier to useโ.
60
36
771
either build a few gpt wrappers with great distribution and run them as cash flow businesses or raise vc and swing for the fences on a hard tech idea. in between is a wasteland.
21
45
710
Someone should start a studio of cracked devs + GTM team and just look at the YC batch every season for the best ideas and then front run them.
54
15
713
Iโm going to start hosting dinners at my apartment in San Francisco for 8-10 people twice per month. If you are a founder, and especially if you are NOT a founder, and you would like to come for an evening of good food and conversation, send me a DM! . First dinner is Dec 5!.
49
11
628
This little piece of startup culture is pretty stupid. Ideas are not worthless. Most people should spend considerably more time thinking about what to work on before they start.
@0xSamHogan Ideas are worthless :).
46
20
498
Bittensor already ripping off @NousResearchโs distributed training algorithm that was first published like 72 hours ago and not even mentioning them in the announcement tweet is entirely on brand.
AI History is being made. Distributed Training on Bittensor. Made possible with @MacrocosmosAI + @manifoldlabs + @const_reborn
38
17
459
Ilya now has billions of MSFT dollars at his disposal without fiduciary duty to turn a profit. If everyone working on commercialization leaves, so what? It only increases runway for real science. This is a dream scenario as a researcher.
54
25
404
Looking for a junior Idea Guy to join the team. No experience required. 70% equity. Youโll be leading a team of 3 world-class engineers to bring your ideas to market. You should have a very small blog or podcast where you already share your ideas that we can review. DMs open.
60
22
410
Companies like LinkedIn store structured user data in a database. They convert that data to semi-structured HTML to display to users. Using LLMs itโs possible to convert semi-structured HTML back to structured data and store in DB. Theoretically you could vampire attack most.
45
34
360
So who is going to raise $500B to build infrastructure for open source models?.
93
22
362
It has never been easier to bootstrap your way to 100k MRR and live a wonder life. This should be the preferred path over venture funding unless you are explicitly going for a 10B+ outcome building world-changing technology.
This was a cool listen. I think Cloud+AI is increasingly making the @levelsio -style model of a scrappy solo serial micro-entrepreneur viable, allowing one person to spin up and run a number of companies that generate income, possibly well into billion-dollar valuations.
3
18
355
Introducing @kuzco_xyz, a distributed GPU cluster for LLM inference on @solana. Kuzco allows anyone with a GPU to get paid for serving inference requests for top models like Mistral and Llama2. Inference Kuzco via our @OpenAI compatible API. Available today for macOS and Linux
78
55
307
๐ I made GPT-4 convert BabyAGI to TypeScript, so @yoheinakajima doesn't have to. Introducing BabyAGI-ts: a CLI tool that can be installed as a global NPM module and used to create and manage BabyAGI agents. Code Here ๐.
9
40
324
Next time youโre trashing a YC founder for a slop project remember he might only be 500 days from exiting for $3B.
8
2
333
๐ AI-powered docs for @vercel, built on @LangChainAI. Ask questions, get answers, view sources, in < 5 seconds. This is super useful for dev communities. We're using the new gpt-3.5-turbo model released yesterday. It's blazing fast, and the answers are scary good. Link๐
20
33
309
Here's a few thoughts on @LangChainAI, the problems I see with it currently, and how I think it could improve. This was originally formatted as a message to @hwchase17:. Here's a few things off the top of my head -- . 1. Heavy use of OOP. Having multiple layers of abstraction.
31
30
314
We built a @LangChainAI-powered UI that can be embedded on any website with just a script tag, similar to Intercom. This is huge for b2b chatbots. If youโre interested in an open-source boilerplate for making something similar, star the GitHub repo in the comment below.
8
34
271
Just had @cursor_ai and Claude rebuild Minecraft from scratch. It one-shotted the game mechanics ๐คฏ. Inspired by @levelsio. Using three.js from @mrdoob. Multi-player coming next โก๏ธ
BUILD A GAME WITH AI!. I'm going to use @cursor_ai and a mix of GPT4.5 and Claude 3.7 from @AnthropicAI to rebuild Smash Bros from scratch. I'm not sure how this will go. Let's find out ๐ค.
27
11
278
A better option: skip the bs, donโt raise money and own 100% of your company from day one.
The most sophisticated founders know this and set up 8 year vesting at the start.
13
5
268
๐To welcome our friends at @helium to @solana, we trained GPT-4 on over 800 pages of Helium documentation. Introducing HeliumBot, by @contextbots; a chatbot to answer all your questions about Helium and the Helium network. Try it out below ๐.
8
31
215
build an open source email client from scratch
Forage Mail (@foragemail) solves email overwhelm. It filters low-priority emails and sends you a clean, daily summary email, including a TLDR for every newsletter. It learns your preferences, supports custom rules, and works with any mail app.
15
6
221
show up to the nyc function. talk to zero women. drink only water. in bed by 10:30
20
2
213
I walked passed the @cursor_ai office last night at 11pm and they were still there working. In fact, every time I walk passed their office they are there working. Every single time. These guys are no doubt gigabrains but they also work hard af.
Weโve trained a new copilot++ model thatโs smarter and much much faster. @cursor_ai takes the crown the fastest copilot. Out next week!
2
7
214
Introducing autodelve!.
Introducing Autodelve ๐คโก๏ธ. YC-backed AI customer support tools quoted us nearly $20,000/yr for a simple Discord Q/A bot. So built our own with @cursor_ai in 2 hours and made it open source. Fork the repo, index your docs site, and get answers in Discord in under 5 minutes
12
7
208
the agents-sdk from @CloudflareDev seems like a good fit. It includes a detailed cursor rules file that should help speed things up
Want to quickly prototype or build an app or agent with Workers + your favorite AI model?. We just published an example prompt you can use to get better results out of Claude, o1, o3-mini, Gemini, and Llama 3.3:
2
5
195
Being a solo founder is good and based if you want to build a comfortable lifestyle business, drive a new Toyota Forerunner, and vacation a few times per year. Otherwise find a partner, or several. Thinking you will build a generational company by yourself is pure hubris.
24
0
177
โจLangChain-powered documentation search for the new @LangChainAI JS docs site. Ask a question, receive an answer with citations. Link in comments. What OSS project should we do next? . P.S. This can be embedded on any website ๐
9
21
156
18 months ago, I left Montana for SF to join the @fdotinc accelerator program. My only regret is that I didn't come to the Bay sooner. Most of the things I struggled to make happen in 5 years of trying outside the Bay happened within about 9 months of moving to San Francisco:.
15
11
158
Asking people at SF parties if they like TypeScript and scoffing no matter how they respond.
12
1
158
We are now serving large models on @OpenRouterAI . - DeepSeek V3.- @NousResearch Hermes 3 405B.- DeepSeek R1 coming soon
7
12
154
over one thousand people DM'd me about coming to dinner. people want to hangout. They want to meet new folks and have interesting conversations, and they want to do it over dinner in a casual environment. let's make SF the city of dinner parties in 2025!.
Iโm going to start hosting dinners at my apartment in San Francisco for 8-10 people twice per month. If you are a founder, and especially if you are NOT a founder, and you would like to come for an evening of good food and conversation, send me a DM! . First dinner is Dec 5!.
9
3
147
The smartest people I know are spending 50+ hours/week building on @solana. Does this mean it will 100% succeed? No. Of course not. But Iโm still all Iโm.
12
24
134
I'm going to use from @firecrawl_dev to create llms.txt files from Discord documentation to give Cursor more info on how the Discord API worker
1
2
141
It is highly likely that the person who eventually re-imagines personal computing to build the AI-native OS for consumers is building viral GPT wrappers right now.
6
5
138
People pay $6,500/month for a 1 BR apartment in SF. Meanwhile, I pay $1,200/month for this in Polson, MT. Remote work has completely changed what's possible.
36
1
126
A 23 year old recent Colombia grad crypto vc researcher just told me that what weโre building wonโt work because โitโs too hard to build a gossip protocol with Solanaโs small transaction size limitsโ. That itโs guys. Weโre throwing in the towel. Canโt argue with facts like that.
22
4
131
Crazy that this guy is worth $200B and is almost single-handedly powering all open-source AI R&D
14
1
133
Like this tweet if you want to come to a banginโ party at our new office.
10
2
129
2
0
127
Join us in-person in downtown SF!
We're hiring engineers @kuzco_xyz !. If you know anyone who'd be a great candidate, we'll pay you a $10,000 referral bonus. DM me or @0xSamHogan to chat.
4
0
124
๐ I made GPT-4 read documentation from the top 34 @solana projects, so you don't have to. Introducing Solana Ecosystem search, by @contextbots. Ask questions about any Solana project's documentation and get immediate AI-powered answers. Link in top comment๐
13
29
114
@realEstateTrent Youโre going to ask 500 people if you can skip passed them? One by one? Group by group? What if someone is having a bad day and says no? Youโre going to fight with them about it?.
29
0
119
@rauchg Big bro single handedly invented serverless computer internet architecture and protocol. Incredible.
3
1
123
I am once again reminding you that structured extraction using LLMs is going to transform every single industry in the next 10 years.
outline's growth is insane, using is an understatement!.
6
8
120
Youโre only fine tuning llama 7b bc itโs easier than finding PMF.
5
11
111
he has risen ($10M at $100M cap, no MFN. led by tiger global).
6
0
112
Come work with us in our sunny downtown SF office!
We're hiring engineers @kuzco_xyz! . If you know anyone who'd be a great candidate, we'll pay you a $10,000 referral bonus. DM me or @0xSamHogan to chat.
10
2
106
GPT-4 is REALLY good at extracting data from HTML using OpenAIโs function calls. Itโs like a super scraper for the whole web โ if the data exists on a page, GPT-4 can grab it. So many use-cases itโs hard to overstate. Iโll be talking a lot about this over the next few weeks.
8
7
109
More interestingly, could build a SQL database for the whole web. Demo launching next week.
8
2
106
the guy who just raised 200MM or whatever to build a closed-source AI chatbot company is now shilling his new book on how open source AI is going to kill us lol.
The last century was defined by how quickly we could invent and deploy everything. This century will be defined by our ability to collectively say NO to certain technologies. This will be the hardest challenge weโve ever faced as a species. Itโs antithetical to everything that.
9
9
100
A bit more about what I'm up to with @kuzco_xyz. TLDR: We're building a distributed compute market for LLM inference that combines inference capacity from discrete providers into a single cohesive network, running at the edge. Full memo here
11
9
100
โจChatGPT-powered search over ALL public @openbb_finance content. This includes:. - OpenBB blog.- OpenBB docs.- OpenBB YouTube channel. Ask questions, get instant answers derived from multiple datasources, with links back to source. Link๐ . cc @LangChainAI @josedonato__
8
17
103
@fuckpoasting Hard technical skills will cease to matter. These are the smart people with a limited window of time. Getting rich off rizz, beauty, or bravery will always be an option.
1
0
97
@CloudflareDev Claude is struggling to find a good way to integrate with the Discord API and keeps suggesting we use React code on the backend. We need to give Cursor more context on how these APIs work.
4
0
99
My girlfriend has a pretty darn good idea for a AI + Enterprise HR product. She has the experience / connections to sell it. If youโre in SF and want to build a RAG app with a pretty clear path to a $20-30M exit in a few years, hit my DMs with repo/url for RAG app youโve built.
18
5
96
Asking people who just moved to SF if theyโve read Zero to One and scoffing no matter how they respond.
16
1
92
The cost of LLM tokens has dropped more than 50% over the last 12 months. @GroqInc was leading the pack for a while, but just isn't competitive any more. Llama 3.1 70B tokens on are now 63% cheaper than @GroqInc and almost 90% cheaper than @replicate
12
6
93
there are clear signs when someone finds their calling
5
4
97
@JoelLoren1 Why not report this to the relevant authorities? If theyโre using illegal labor itโs easy to prove.
23
1
83
Yeah Iโm going to code more bc of ai, not less. ai is just a way to increase bandwidth between programmers and computers. donโt over think it. yap less, ship more.
8
5
92
We failed to return responses to over 11,721 Llama3 inference requests since 8AM this morning. This represents about 0.63% of the 1.7MM requests we handled in total. The Kuzco API is free until we're at 99.99%.
Llama3 inference powered by ~2,000 globally distributed GPU nodes. Available now. Use the network in 20 lines of code. Example repo in comments.
11
7
80
Visualizing GPU utilization during LLM inference batches on the @kuzco_xyz distributed network in real-time. Repo + blogpost coming soon
11
11
81
I legitimately need solana transaction fees to be $0.00000001.
Not sure people have understood today. Frankendancer did not just run a benchmark on a single machine and got like 50k tps. They ran 100 commodity machines that are globally distributed at 1M tps. 1M. Not colocated . And not even full firedancer. That is fucking insane.
7
3
79