Bede Carroll

Network Engineer, Automation Lover

[Bede Carroll @BedeCarroll]

@redtachyon With agent-agent and agent-human costs I’d say there’s actually a detriment to having high human-human workflows as well. I don’t think it’s there juuust yet though. Organizational structures really haven’t caught up yet

[Bede Carroll @BedeCarroll]

If I had a guess this is what happens when you start to get good at something. If you play ranked games you’ll notice “clicks” and jumps in performance. You start not thinking through actions and just have impulses

[Tenobrus @tenobrus]

balatro neuralese

[Bede Carroll @BedeCarroll]

Bro is out of distribution 😭 ngmi

[Daniel @growing_daniel]

Eight sleep uses AI to guess the temperature you’ll want and it’s somehow wrong literally 100% of the time

[Noam Brown @polynoamial]

https://t.co/oWqzT12RtZ

[Bede Carroll @BedeCarroll]

@flowersslop You can use an automation for this in codex. I have something similar for a bunch of work related tasks

[𝗥𝗬𝗔𝗡 𝗦𝗧𝗢𝗥𝗧𝗭 @withzombies]

. @_lopopolo

[Bede Carroll @BedeCarroll]

@RhysSullivan I don't even think it's the right thing to do anyway. The models behave differently and need different instructions

[josh (oldfriend99) @oldfriend99]

If ai is not conscious then explain how the mcdonalds touchscreen station knew i wanted an extra cheeseburger

[Bede Carroll @BedeCarroll]

@hd_nvim Have you made language choices based on this out of interest?

[Matt Stockton @mstockton]

Stepping away from the dopamine chat box and recording a long, rambling, yet intentional voice memo covering everything you want to accomplish in whatever project you’re working on. Then transcribing that and using it as input to the model, where you then ask the model to ask you questions exhaustively until it knows exactly what to do in order to accomplish your goal

[Bede Carroll @BedeCarroll]

I think we need to understand that the“evil Martian colony administrators” in the movies were simply under crushing valuation and financial pressure

[forward deployed ccp gf @FangYi11101]

Valuation guys trying to discount the probability-weighted Mars colony scenario.

[WAV Archive @WAVArchive]

Tristram – Matt Uelmen (1996)

[Bede Carroll @BedeCarroll]

Seriously underrated. When friends tell me about compaction issues with competitors I struggle to understand it because it’s so good in codex. The post compaction error rate is very low for me to the point I have single sessions with +++ tokens

[labib @labibrahman]

codex compaction is arguably the single biggest user experience improvement in ai in the last 6ish months and does not get as much hype as it deserves

just don't care about context window anymore and somehow it does "understand" the whole context

[OpenAI @OpenAI]

Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.

For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.

An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.

This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.

[Thomas Bloom @thomasfbloom]

An internal OpenAI model has disproved one of the most well-known Erdős problems: the unit distance problem.

This is, without doubt, the most impressive achievement of AI in mathematics so far.

https://t.co/J0duJXNbph

[Bede Carroll @BedeCarroll]

I tell the same story when someone asks for my updates during standup

[JT @jiratickets]

asked my grandpa what he thinks about the new spotify logo and he told me he saw a man get run over by an M4 sherman tank and pulverized into jelly in korea. the rain diluted his remains into the mud and his existence was erased from history in just a few minutes

[Bede Carroll @BedeCarroll]

This will be a fad. As ICs become more productive with AI coordination problems get harder. Until AI automates the stack of course..

[Mo @atmoio]

"I don't think people managers will have any value in the future."

Airbnb CEO said on a recent pod the people most screwed by AI will be people managers.

Perhaps we had this all wrong. The future belongs to ICs. There is now more use for them than ever before.

[kei @thelissimus]

LLVM! Did you delete the bounds check?!

[Bede Carroll @BedeCarroll]

I have no idea on what the future of personnel for software engineering looks like but I will say the coordination penalty becomes more acute with LLM use

[Bede Carroll @BedeCarroll]

The numbers are so low. Only $10k in tokens for an employee? I would feel so constrained

[elie @eliebakouch]

Arthur Mensch answers to the french representatives:

"our (mistral) models are capable of finding all the vulnerabilities found by mythos"

"There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed"

some numbers:
> 1B R&D spend at Mistral this year
> at Mistral 10% of salary mass is spent on tokens
> estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$
> 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value.
> revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France.
> a bit less than 30% of Mistral capital is held by US VCs
> Mistral's goal is 1GW in 2029
> they train/will train bigger models internally and distill them to serve to customers
> Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year

some of their current clusters:
> 40MW in France
> 25MW in Sweden
> 80MW in France (next year)
> they train models on "10s of MW", mention that they need the gpu to be collocated to train model
> insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint