Ian Butler (@kinglycrow) header
Ian Butler (@kinglycrow) profile picture

Building https://teapawt.org/t.co/sOIl0f99Oi. AI Researcher @XBOW, thoughts my own. @aitinkerers SF co-organizer https://teapawt.org/t.co/9y4zk8oY83

San Francisco, CA
Joined May 2022
Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)Photo by Ian Butler (@kinglycrow)
Your users aren't going to turn AI into a hobby. The product has to bring the capability to them. Great doing your first podcast episode with you both @tbrownio @ParsaKhaz thanks for having me! Follow them to see when the whole thing drops.
2
1
14
3,021
Gee I sure hope the new OpenAI models are good! *looks at timeline* Oh. Oh no. I really hope they’re not as bad as I see a few people saying and that it’s a coordinated campaign by Ant because otherwise ☠️
1
79
My concerns about Anthropic can’t be understated. The worst things are brought about by people with the most righteous interpretation of their actions.
Now we can make sense of why Rubio repeatedly refused meeting requests from Jensen.
1
6
301
I used to spend so much time filling out forms for medical stuff, moving stuff, various applications and what not and now I just tell Nym to do it because it has all my information and the little time savings really adds up over time.
1
104
I am deeply interested in multiplayer collaborative environments for agents of all kinds. I cannot wait to share my take on this in fact.
software factories feel like a temporary problem. the models will learn to build them for you the more interesting question to me is how to build a workspace where agents can collaborate with you
2
203
We love synthetic data in this household.
1
109
Ian Butler retweeted
Replying to @vikktorrrre
“When a model is released in an uncontrolled manner, there's no ability to do that. It's entirely out of your hands.” Yeah Dario, that is EXACTLY the point. We want our life / business to be out of your hands.
7
25
555
13,475
A really cool thing to be able to run on your own hardware is PII anonymizing models and they've gotten much better. Super useful for helping properly handle customer data for analytics etc.
99
I mean hey that's how you get more performance out of me 🤷‍♂️
I wonder if you could get better performance in any ways (e.g., writing style, original solutions) with LLMs if you let them be much more disagreeable, much more impolite, much less reliable
1
4
267
usenym.com has all the inline generative flourish you could want.
OpenAI’s Dot is capable, but seems to be falling further behind in its interface compared to Muse and Duet. The single-threaded conversation, lack of inline generative flourishes (like maps), & the inability to see what tasks it is assigned makes it harder to use than the others.
1
168
Every time I go to a demo night in SF I walk away feeling not particularly impressed like 80% of the demos didn’t work. I don’t need like ground breaking work but have some pride and craftsmanship. The demo god excuse is played out.
2
8
780
Yes it’s hard to get up and share stuff, no I won’t applaud someone for just doing that. I want something out of my two hours. In the age of AI simply building something isn’t that impressive.
1
99
As someone who hosted and curated events for over a year I get you get slammed with a bunch of crap and the work to separate out good from bad is quite asymmetric, but like I literally saw someone demo something tonight that was broken in the exact same way when he demoed it at an event last month. Theres gotta be ways to keep that shit out.
1
88
I've heard a lot of feedback that people like when agents feel a little more playful, so Nyms can now react to your messages with emojis! Not a crazy feature, but a little polish that I hope people appreciate.
6
198
Okay today is the day I finally escape LastPass. For years they've been just under my threshold for being too annoying but today is the day. @1Password here I come.
3
7
437
I loved adding these to Nym btw.
Nyms can now react to your messages with an emoji: a heart, a thumbs up, a laugh, an exclamation, a question mark or a thumbs down.
3
164
I just got a text about it reaching out to correct a service's pricing for me they mis quoted their pricing and Nym caught it for me and made me aware before I had a chance to let it slip through.
2
230
Okay this is what I'm saying. I grew up in a finance household, my mother was a senior director. The stuff she told me went on in the places she worked this isn't even that crazy.
anyone who has an issue with this wouldn’t survive a day in finance
6
711
I also don't want to like yuck anyone's yum here. I've just been building harnesses of all types for 4 years now and my take is the ceiling on intelligence is the model. Harnesses can make things faster and cheaper, provide more coverage, in short they help expose the capabilities of the model in different ways and then you can definitely elicit better outcomes by learning to use the tools better. BUT that means since the cap is model intelligence, after you reach a reasonable level of proficiency you're just asking for diminishing returns by reading into everything everyone says. We're in the VS Code era for agentic programming, which is to say there are established major players with good systems and other competent niche systems, but you can basically pick the one that feels the best to you, get some reasonable skills installed and then build that proficiency. Once you do that you're going to be getting say 90% of the value and over investing into the discourse is going to distract you from building the actual things you want to build. It becomes a distraction.
I've been pretty quickly distancing myself from the overall coding agent discourse because frankly I don't think there's a lot of rigor in it it feels like the JS framework era but like 100x worse.
14
4
3
86
8,767
I've been pretty quickly distancing myself from the overall coding agent discourse because frankly I don't think there's a lot of rigor in it it feels like the JS framework era but like 100x worse.
17
20
3
397
20,257