@mattgillard looks very cool! thanks!
Notes
- @Shivipmp they could build it themselves but would rather purchase an abstraction so they can focus on what matters to them
- @jerryjliu0 @shijith @llama_index thanks! I'm looking for a place to upload pdfs and then provide a tool to LLMs. does that make sense? https://x.com/stevekrouse/status/2024192763424637063?s=20
- @intellectronica thanks! i'm looking for a guide or tutorial that shows how you add it into an llm's list of tools for example, the attached exa docs is what i'm looking for, but for pdfs https://x.com/stevekrouse/status/2024192763424637063/photo/1
- I have a customer with a ton of PDFs they want an LLM on top of, but we're hitting context window limits Is there a high-level API that lets me upload a bunch of PDFs, and then provides a "tool" that I can give to an LLM?
- @partlysunnyai @kaseyklimes kasey and i do get coffee! thanks for the thought :)
- http://blog.val.town
- new blog design! https://x.com/stevekrouse/status/2024130434003513703/photo/1
- @andy_matuschak @geoffreylitt Seems like it's just buying 18 months into the future. If current trends hold, everyone will get opus 4.6 fast intelligence for pennies in a year or two But by then you'll want Opus 6.6 Fast...
Text may be partial · Read original
- @geoffreylitt ha! do you know your token budget?
- @geoffreylitt exciting! is this in claude code? how much $ are you spending on opus in a day? i've probably let 10x prices scare me away irrationally. buying one's way into the future is money well spent
- @lastgoodhandle that's what i mean! learning how to think and learn is the hard part actually producing lines of code (by moving your fingers on a keyboard) was always the easy part
- RT @simonw: Short musings on "cognitive debt" - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lo…
Reposted by Steve
Text may be partial · Read original
- @aboodman Agreed! People don't seem to care about this part
- Today I learned that Geoffrey Hinton's great-grandfather was George Boole! https://x.com/stevekrouse/status/2022749771224461763/photo/1
- @twschiller Yeah we don't even need a filesystem. And yes playwright MCP would work great
- @MatijaSosic @MatijaSosic last night I dreamed that you replied to this tweet with: "in other words, llms should use libraries" and i thought it was a banger!
- "Writing the code was never the hard part" It has never been the bottleneck. The bottleneck has always been human understanding. Reading code, not writing code. https://x.com/DavidKPiano/status/2022670822482071790
- Are there long-running async coding agents (like codex, claude code) that you can point at an mcp server instead of a github repo? In other words, my users want to use codex cloud with val town, but I'm not able to figure out how to connect it to val town I imagine agents would LOVE coding in val town, because the feedback loops are so fast, save, deployed, view logs, view live website, iterate till done We could build a long-running Townie mode ourselves, but I'd love to not have to rebuild something that already exists...
- @jdan 💯
- @simonw This result makes me so much less excited about this model
- @brettberson @X you're begging someone to go back through your tweets and pull out the good/bad predictions. i may do it! i hope you tweeted something about how devtools is not good vc-investable category ;)
- i love the large text paste as attachment ui https://x.com/stevekrouse/status/2022412944047509629/video/1
- 1. my friend and i agreed on a time to chat 2. i sent a cal invite 3. i saw he sent a cal invite 4. i deleted my cal invite 5. he also had seen my cal invite 6. he also deleted his cal invite 6. we both apologized 7. we kept looping like this for the rest of the day race conditions are hard
- shipped mermaid diagram support in val town readmes! thanks @dthyresson for the feature request! https://x.com/stevekrouse/status/2022385012881178715/video/1
- @jonathoda Thanks!!!! Yes, a lot of these tricks are interim phases. And lots of people will be able to build complex apps in English without learning anything about code. Which is exciting! But the "Real Software Revolution" is when these tools make you smarter as you use them
- pleasantly surprised claude lists us first here! https://x.com/stevekrouse/status/2022372647460393164/photo/1
- @danshipper 👀
- @danshipper i want to watch you play with it today!
- RT @MatijaSosic: "Swarming 20 agents is the equivalent of wondering how in 1910 you'll hitch 20 horses to a wagon to achieve 60mph speeds.…
Reposted by Steve
Text may be partial · Read original
- @MatijaSosic ❤️
- @MatijaSosic Yes and folks were upset about the spam
- @MatijaSosic We tried this and it went poorly 😅
- @MatijaSosic https://x.com/i/status/2022317376046240196
- @fw3d @ValDotTown Woohoo!
- @rileytomasek Great question! No, delegation and management still are inferior INTERFACES no matter how smart the agents gets Spotify Discover Weekly is a good example of a beautiful interface to lots of intelligence I don't have to manage But it's a black box.We could improve the tool…
Text may be partial · Read original
- @MatijaSosic https://x.com/i/status/2022317376046240196
- @MatijaSosic That intermediate representation would just be code at a higher level of abstraction, like wasp!
- @MatijaSosic That intermediate representation would just be code at a higher level of abstraction, like wasp! https://x.com/i/status/2022317376046240196
- The trouble with distraction is that you are too distracted to know how distracted you are The trouble with outsourcing your thinking to LLMs is that you're becoming too dumb to realize how dumb you are



