James (@wattersjames) profile picture

“There are some things which cannot be learned quickly, and time, which is all we have, must be paid heavily for their acquiring.“ —Hemingway

San Francisco
Joined April 2009
For those who aren't aware: you write an OpenAPI manifest for your API, use human language descriptions for everything, and that's it. You let the model figure out how to auth, chain calls, process data in between, format it for viewing, etc. There's absolutely zero glue code.
56
478
219
4,154
1,033,577
I've developed a lot of plugin systems, and the OpenAI ChatGPT plugin interface might be the damn craziest and most impressive approach I've ever seen in computing in my entire life.
80
679
78
5,717
1,619,101
“The most courageous decision that you can make each day is to be in a good mood.” — Voltaire
7
242
19
1,241
83,913
People are too focused on code generation and completely ignoring that LLMs are useful for code analysis. I've been personally surprised how useful they were in identifying missing test cases, unreleased leaking resources, or even telling me what's wrong with my IAM policy.
16
59
7
606
128,369
now reading
People have been asking if we can automatically generate API tests by looking at API traffic. This is now possible with our new AI chat dog, Aki! 🐕 Read more about Aki in this blog post: akitasoftware.com/blog-posts…
1
3
2,415
Notable
I haven’t programmed full time for something like 12 years now. Yet, assisted by copilot and ChatGPT, I’m more productive than I would be if I never stopped and wouldn’t use them. A heartfelt thank you to the arc of progress.
842
Imagine if the improvement curve here resembles anything like the evolution from Apple II to modern PC I grew up in
1
1
3
1,536
Okay, I was initially skeptical but our new fancy lil guy has exceeded all expectations when it comes to summarizing API behavior. 🧐 Aki is surprisingly smart! (And ChatGPT is surprisingly good at technical things.)
Users have been asking about whether Akita can help with documentation, test case generation, example payloads, and more. Meet Aki, our exceedingly polite, ChatGPT-powered AI dog who can help with many of your API needs. See examples: akitasoftware.com/blog-posts… Try us out today!
2
4
23
4,887
GPT-4 is waaay better at programming than given credit for. HumanEval is a benchmark of python programming problems. With some prompt engineering, GPT-4 scores ~85%, destroying Codex's 29% from just 2 years ago And performing much better than OpenAI's publicized accuracy
8
51
10
409
99,508
Replying to @TheMarkONeill
Yep! Platform teams are absolutely starting to trend. I see this in my work, though my view is limited compared to yours! Stream-aligned teams were arranged to facilitate fast production. But now there’s inefficiency in duplication. How do we resolve this? Platform teams! – at Scotts Valley, CA
1
1
5
800
Looking at our inquiry trends on platform engineering. Holy moly. What a rocket ship of a topic. – at Massachusetts, USA
3
2
18
1,720
James retweeted
“Movement is my habitat. My only rest is motion. … Let me go where I have not yet arrived.” — Luce Irigaray, Elemental Passions
6
49
3,274
How much would it cost to train GPT2-1.5B from scratch today? Original was estimated at $43K but we have four years of hardware and algorithmic advances since then, right?
12
6
2
149
44,504
Go SF!
Excited to see everyone at The Commons in SF tonight starting at 6 🤗 Based on the austin/seattle meetups we expected ~20, and rented a space to support 80 people. We currently have 450 people who have RSVPd 🤯 I’ll be at door to personally apologize to people we can’t let in 🙏🏻
961
Excited to see everyone at The Commons in SF tonight starting at 6 🤗 Based on the austin/seattle meetups we expected ~20, and rented a space to support 80 people. We currently have 450 people who have RSVPd 🤯 I’ll be at door to personally apologize to people we can’t let in 🙏🏻
4
1
4
90
85,090
Wild numbers from the latest GitHub Copilot blog post 46% of all code written 🙃
53
144
48
2,001
849,629
Cool example of its limits
Replying to @JoINrbs
ChatGPT plays correct moves maybe half the time, plausible moves 90% of the time, and illegal moves (eg sensing its queen through other pieces) 10% of the time. It’s memorized openings, so it starts strong and goes downhill from there. Good reminder to not trust but verify LLMs.
3
1,557
One thing I really like about language models is that they are stateless (they are functional programs of type text -> text). This allows us to share prompts (essentially currying the LM) and reproduce results.
10
69
6
304
104,584