Adam Smith
@asmith
Followers
5K
Following
1K
Media
65
Statuses
3K
Founded Xobni and Kite, MIT, YC 2006, VP Eng Affirm; https://t.co/FEmC9AfHuf https://t.co/SlPMsPZ5xm https://t.co/Uu6jnPYRuR
Joined April 2007
I want a factory now, ngl
129
105
3K
Stop guessing Incoterms. Grab our free quick-reference sheets and know exactly who owns cost + risk on every move. → Click for free download
10
35
145
THE FOUR STYLES OF CONFIDENCE ON A TEAM I think of everyone as having a property: how strongly they state their opinions, divided by how right they actually are. There are four common buckets. People with high ratios are OVERCONFIDENT. They always pound the table, even when
1
0
2
Every teammate has a confidence ratio: how strongly they state their opinions, divided by how right they actually are. The 4 styles—and why proportional confidence wins. 🧵
1
1
4
Source: Trait and facet personality similarity and relationship and life satisfaction in romantic couples, Supplementary Material
0
0
1
Earning shouldn't disappear behind a paywall. Earn 2% APY on $USDG with Kraken. No subscription required. Power users can earn more with Kraken+
3
12
70
Female vs Male life satisfaction with varying levels of male-female partner agreeableness. Many interesting conclusions here
2
0
1
In U2’s 2009 album, Bono says “The best of us are geniuses of compression.” That’s something one LLM might say to another 🤠
1
0
0
Watch @typesfast pitch Flexport to Paul Graham and Sam Altman for the first time at Startup School 2013:
58
79
2K
OK, say it goes further and we get AI that's broadly superhuman across all domains. How do you invest for this? Does the question even make sense?
5
1
8
This is one of the most important results of the year, as exciting as the OpenAI o1 results. We now have two very different, very promising scaling dimensions for test-time compute: TTT and RL. (1/🧵)
Why do we treat train and test times so differently? Why is one “training” and the other “in-context learning”? Just take a few gradients during test-time — a simple way to increase test time compute — and get a SoTA in ARC public validation set 61%=avg. human score! @arcprize
1
2
11
Sounds in language often correlate with meaning. For example, all of these words relate to the nose: sniff, snout, snot, snore, sneeze. LLM embeddings place semantically-similar words together It would be fun to study the similarity of syntax and semantics using LLMs 🙂
1
0
6
Seems about right: 3 or maybe 4 orders of magnitude from increasing investment, 1 from Nvidia margin compression, 1 from custom silicon, 1 from Moore's law. So plausibly 6-7 OOMs of headroom in the current sprint, which would last until the early 2030s. (Absent a wildcard on
Given that you need 100x more effective compute between model generations, if we don’t get AGI by GPT-7, will we just never get it? @_sholtodouglas: “GPT-4 costs, let's call it, $100 million. The $1B, $10B, and $100B run, all seem very plausible by private company standards. You
6
1
18
Maybe a counterargument could be that novel math requires more creativity, but LLMs can brute force that
0
0
3
Which will be harder for AI: programming, or mathematics? My guess is programming. Both have to be exactly correct, but programs are longer with more non-local context.
2
0
4
The most educational tweet of the year
language models just being programmed to try to predict the next word is true, but it’s not the dunk some people think it is. animals, including us, are just programmed to try to survive and reproduce, and yet amazingly complex and beautiful stuff comes from it.
0
0
1
Why is AI amazing at art, but can't drive a car? Can we predict which new AI applications might arrive soon, and which are still years away? New blog post thinking this through 1/n
2
2
25
Manually encoding knowledge into AI systems is never the long-term solution. At Kite we connected our Python static analyzer to ML models using Graph Neural Nets. It may be a good short-term solution but won't be the long-term approach https://t.co/TPFXvxVaFT
0
0
2