@__drewface

Happy husband and father. Trying to make the world a teeny bit better each day with my family and friends. Teaching people to love the new era of computing.

New York, NY
Joined September 2013
I haven't tweeted much about it, but I am running a part-time, 6 week program this summer at Fractal to help professionals master AI agents. It's based on everything I learned from training ~100 engineers over the last 2 years. Check to see if your employer will pay for it :)
4
4
3
113
16,742
RT @catehall: With effective altruism in the news, I thought today was as good a day as any to finally take the Giving What We Can pledge h…
1
14
I'm really enjoying Cuties (the new social media app created by @christineist). The only problem is that all my followers from Twitter aren't there. Please join me on cuties and comment on all my posts so we can become friends!
1
2
1
13
858
This is very good work, and one interesting thing is that I think basically anyone could have done it. If you think "there is nothing I can contribute to AI research", think again! This is a new field.
Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.
1
13
677
The internet is a real place, and it costs you nothing to encourage someone, but it costs you enormously to be mean. At the end of it all, at the end of our lives, I really believe we will be judged for every time we were mean or discouraging in an internet comment.
This quoted post is unavailable.
1
3
238
>be me >open robinhood after getting a letter saying they are going to send all the money to my state because my account is inactive > i was invested in AMD > 590% return > i only had $250 principal, smh > last sale on the account was $1000 of NVIDIA (2021) > 1800% return missed
3
149
6,246
Instead I sold all those stocks in order to be an entrepreneur lol
2
27
952
anyway, i wouldn't change a thing
3
29
817
Andrew Rose retweeted
Jev can generate game levels in real time. Faster and cheaper structured output could be a big deal for game dev!
126
210
117
2,643
462,622
select all squares containing a reason we are doing this
2
3
29
666
It's strange to me to use numbers (the concepts we invented to communicate quantitative precision) when giving intuitive judgments. You can't supply the kind of precision that numbers are meant for, so you're using the numbers in p(doom) as something other than quantity. What?
3
6
546
I'm not really against the idea of having number-like entities for this. I understand why it's useful to be able to communicate "my risk assessment is more or less than his/hers".... but why a 100-point scale, then? Why not 1000 or 10 or 5?
3
1
162
One thing I teach my students about is Specificity in prompting Agents. Compare these 4 prompts: 1. Make me a good website 2. Make me a flappy bird clone 3. Make me a perfect clone of Flappy Bird 4. Make me a perfect clone of Flappy Bird, using Expo and the Expo simulator
1
2
23
972
One way to create Specificity in a design is to create a document called a Specification, but this is a rather circular definition. There is nothing special about a file called SPEC.md. The special property is in the way our words constrain possible interpretations!
1
1
9
317
This is why I don't use approaches like "spec-driven development" which rely on agents to create a spec based on a user interview. There's nothing wrong with that, but I believe practicing good specification allows us to specify HUGE amounts of complexity in tiny prompts.
3
7
283
what do you wish for, user?
🤖 Made with AI
2
1
1
18
840
anyone have strong opinions about the best possible "default" boilerplate for an Astra agent? Django? Elixir? NextJS? Just ask Astra to ship it, and don't worry about it? Do you clone some specific project and then strip it out? Do you specify database?
8
13
1,279
Andrew Rose retweeted
Introducing the US Gov Graph - a complete map of the people and positions of power in the federal government. To fix our institutions we must understand how they work, so we are using AI to model and monitor every government in America. This is Palantir for The People. Check it out here: graph.civlab.org/us We exist to set the conditions for an era of civic excellence unmatched since the Founding. We will need nothing less to navigate the age of AI. I'm looking for exceptional engineers and designers to work with me in San Francisco. If that's you, inquire here: tally.so/r/KY97RK. Self-governing civilization is counting on you.
476
1,546
256
9,373
530,296
I suggest grounding conversations about [doom] in physics. For two reasons: 1. physics allows us to calculate the real implications of mathematical claims (of scaling, mainly) 2. physics is a system which constrains the imagination only to what is allowed by natural law.
1
4
318