Luís Pires @lumpyres

AI Operator | +100k saved via Ai systems Co-Founder https://t.co/9OvRA8TJic @margemx.lab AI systems for hospitality Portugal Joined June 2024

Tweets

37
Followers

78
Following

2K
Likes

117

Luís Pires @lumpyres

2 days ago

Europe is completely missing out on the AI agent coliseum. We’ve got @bot being chaotic, @NousResearch and @openclaw throwing hands, and @Muse is just sitting at the border waiting for a visa approval, likely related to EU AI Act and GDPR.

1 0 0 52 0

View Details

Luís Pires @lumpyres

a week ago

@nateherk 👑😂

0 0 0 47 0

View Details

Luís Pires @lumpyres

a week ago

@Gonzohall @martindonadieu +1

0 0 0 0 0

View Details

Luís Pires @lumpyres

a week ago

Human jobs will rise from execution to arquitects/reviewers in the next months.

0 0 1 24 0

View Details

Luís Pires @lumpyres

a week ago

@adamshuaib This totally matches the pattern I have seen too

0 0 0 735 0

View Details

Luís Pires @lumpyres

a week ago

@al_antdp 100%

0 0 0 121 0

View Details

Luís Pires @lumpyres

a week ago

@DarioAmodei Asking tier 1 labs to unite, and the US gov to "police" the other players that won't join, while only pacing slightly slower to not lose advantage is genius

0 0 4 2K 0

View Details

Luís Pires @lumpyres

2 weeks ago

This is ok coming from architects and why they shouldn't pretend to be structural engineers. Aircraft fuel burns at high temperatures and crash conditions can easily go over 1000ºC. Structural steel softens at half of that (loses ~50% yield strength). In practice load capacity is reduced by a lot, enough to make that happen

0 0 2 2K 0

View Details

Luís Pires @lumpyres

2 weeks ago

@MatthewBerman @bot @SpaceXAI I don't have 100 folllowers but a great Grok Bot use is the to simply enable the team to be more productive and outsource manual repetitive work even through their phones

0 0 0 16 0

View Details

Luís Pires @lumpyres

2 weeks ago

@pmddomingos They want competition to be stronger

0 0 0 37 0

View Details

Luís Pires @lumpyres

2 weeks ago

@systemdesignone Managing agents that are better engineers than I ever became

0 0 0 130 0

View Details

Luís Pires @lumpyres

2 weeks ago

@weswinder Have you tried to change the output style and some prompt engineering? Had great results with both

0 0 0 1 0

View Details

Luís Pires @lumpyres

2 weeks ago

One of the favorite solutions among client projects have been linked to finding expensive mistakes instead of automations.

0 0 1 38 0

View Details

Luís Pires @lumpyres

2 weeks ago

@ben_burtenshaw Great work

0 0 1 60 0

View Details

Luís Pires @lumpyres

2 weeks ago

If slowing the AI labs that take safety most seriously doesn’t stop the race and instead helps the least responsible labs win, why do it?

0 0 3 74 0

View Details

Luís Pires @lumpyres

2 weeks ago

If the goal is genuinely to minimise existential AI risk, under what assumptions does weakening or slowing the most safety capable frontier labs make humanity safer, rather than simply transferring technological leadership to actors with weaker safety standards and constraints?

Anna Wang @a_nnawang

2 weeks ago

I worked at Google DeepMind and now at Anthropic. This is a common sentiment amongst my peers. (I write this in personal capacity.) There is not yet a viable scientific plan to solve risks from recursively self-improving AI. Please look up!

871 4K 23K 2.4M 4K

0 0 1 42 0

View Details

Luís Pires @lumpyres

2 weeks ago

What if the biggest AI safety failure is not racing too fast, but letting the least safety prepared actor win the race?

Jacob Coxon @hilbertspaess

2 weeks ago

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

21K 167K 803K 173.6M 277K

0 0 0 13 0

View Details

Luís Pires @lumpyres

3 weeks ago

@guanlan @guanlan agree on the rest, great work! With this my take would be that Pi becomes the best multi-model/cost efficiency > Codex CLI the best plug-and-play SWE > Claude Code the best strictly on native Anthropic

0 0 1 23 0

View Details

Luís Pires @lumpyres

3 weeks ago

@guanlan @guanlan Pass rates are statistically indistinguishable at 30 tasks with one attempt each, so the ranking might not be fair, right? The 17x cost spread is real, but CC costs comes from running it on Kimi K3, where its caching mechanism does not apply, right?