Fergus Argyll

Karma: 25

Fergus Argyll’s Shortform

Fergus Argyll14 May 2026 21:54 UTC

1 point

2 comments1 min readLW link

Fergus Argyll 14 May 2026 21:54 UTC
20 points
0
on: Fergus Argyll’s Shortform
A quarter of the 13,000 federal lobbyists in Washington are involved in A.I. issues, up from 11 percent in 2023
https://www.nytimes.com/2026/05/13/technology/ai-lobbying-washington-openai-anthropic.html

Fergus Argyll 7 Jan 2026 17:24 UTC
3 points
2
on: Claude Wrote Me a 400-Commit RSS Reader App
Software engineering experience was useful
I think software engineers underrate this. A person who never was interested in computers (programming, or configuring linux, or even studying CS from a book etc all count here as computer experience) has so much less background tacit knowledge. They wouldn’t know how to ask it to make a native app. They barely realize that google.com and chrome are different things. It really helps to at least have a mental model of “There’s a backend with a database, there’s a frontend, there’s an API” etc

Fergus Argyll 13 Aug 2025 11:32 UTC
1 point
0
on: The Problem
Adding ten times as much computing power to an AI is sometimes just a matter of connecting ten times as many GPUs.
Which no one does anymore because it doesn’t work...

Fergus Argyll 8 Apr 2025 17:14 UTC
3 points
1
in reply to: SorenJ’s comment on: AI 2027: What Superintelligence Looks Like
I think he was saying:
By the time the new chip is ready, that will be 1.5 years which implies 5x growth if we assume 3x per year. So; by the time OpenBrain is ready to build the next datacenter, we’re in middle/late 2026 instead of beginning of 26.
Aside from that, the idea that investment will scale proportionally seems like a huge leap of faith. If the next training run does not deliver the goods there is no way softbank et al. pour in 100B.

Fergus Argyll 8 Apr 2025 2:10 UTC
3 points
0
on: AI 2027: What Superintelligence Looks Like
I have a question that I didn’t see anyone ask, but I don’t frequent this site enough to know if it was mentioned somewhere.

Are we sure there will be a 2-OOMs-bigger training run at all?
After the disappointment that was GPT 4.5, will investors give them the 100B (according to TFA) they need for that? In general I’d like to see more discussion about the financial side of the AGI Race. How will OpenBrain get the funding to train Agent-4?
I’ve been looking for markets on manifold to bet this and I couldn’t find a good one. I would bet we don’t get a 2-OOMs-Bigger model until at least 2027 by which chip costs come down enough. My prediction would be OpenBrain etc would focus on fine tuning / wrappers / products / ux /ui etc. for the next 2-3 years.

Fergus Argyll 29 Feb 2024 11:17 UTC
1 point
4
in reply to: lemonhope’s comment on: Supposing the 1bit LLM paper pans out
If I understand correctly (I very well might not), A “one bit LLM” has to be trained as a “one bit LLM” in order to then run inference on it as a “one bit LLM”. I.e this isn’t a new Quantization scheme.
So I think training and inference are tied together here, meaning; if this replicates, works, etc. we will probably have new hardware for both stages