Paul Mallach

Building stuff. Financial markets and technology. Raised on Apple and Commodore home computers, shuffling bits around. Also a sailor.

[Paul Mallach @toolbar23]

finally under way again.

[Paul Mallach @toolbar23]

@redtachyon also make the file big. really big.

[Paul Mallach @toolbar23]

@code_star you don't have access yet?

[Paul Mallach @toolbar23]

@AI_StefanS linkedin? second door down the corridor.

[Paul Mallach @toolbar23]

@4k0n1u5 @cloneofsimo OpenAI and Anthropic are distilling human input the took from the internet and other sources at scale. Why is that more legitimate?

IMHO both are fine.

[Paul Mallach @toolbar23]

@thekitze If it wasn't for the incredible daily traffic jams around Antwerp or Brussels driving in Belgium could be nice.

[Paul Mallach @toolbar23]

@PeteHegseth @DeptofWar @AnthropicAI Anthropic is welcome in the EU.

[Paul Mallach @toolbar23]

@txgermanbre opus and sonnet have one killer feature: turning every reply into a sycophantic love letter to yourself.

when i need to get actual work done, i use codex.

[Paul Mallach @toolbar23]

@WohlstandsWal dumbest post ever

[Paul Mallach @toolbar23]

@txgermanbre Many anthropic key people are supposed to be e

[antirez @antirez]

What Europe should do right now:

1. Call all the European researchers working on AI and return them back with same salary (or they can stay but switch career).

2. Fill EU places having GPUs with money, and put those people there.

3. AI partnerships with China + India.

[Paul Mallach @toolbar23]

@Scobleizer TAM for anthropic or openai cut by 75 percent.

[Paul Mallach @toolbar23]

@Scobleizer TAM for anthropic or openai cut by 75 percent.

[Paul Mallach @toolbar23]

It's interesting to think about how the US export/usage limitations will influence available capital for US labs.

I am pretty sure thtat open models will be at the level of what we get with the allowed models like Opus 4.8 and GPT 5.5 soon, which will make those prices go to zero.

Now the US greatl limits the TAM for all better models created in the states. Less TAM, less capital.

[Paul Mallach @toolbar23]

@TweetsOfSumit @vonderleyen owning a few rtx 6000 pro as a fallback looks more interesting than ever.

unfortuntely unused capital as i would prefer my 200 euro codex sub to even the best current open-source model. renting just a few gpu out while not using them does not seem make too much sense as well.

[Paul Mallach @toolbar23]

Apple's claim deliberately blurs the line between the trusted-agent-thing and an 18 month rollout for that solution.

I am pretty sure, that not the generic, trusted agent idea itself was rejected, but taking 18 month to offer a solution and getting a lock out of competitors in the mean time.

[Paul Mallach @toolbar23]

who will be the phil zimmermann of ai?

[Paul Mallach @toolbar23]

@txgermanbre just make sure whatever happens on the boat, stays on the boat.

[Paul Mallach @toolbar23]

@jpschroeder anthropic has been benchmaxxing for some time.

in benchmarks opus/sonnet always fared much better than in my personal experience.

codex 5.5 xhigh fast ftw

[Paul Mallach @toolbar23]

@AuerDirk btw, did your ai mix up where the 18 months were supposed to be mentioned when writing this, or have talks between apple and the eu about this really taken 18 months?

[Paul Mallach @toolbar23]

sorry, but your article is bullshit. you are just repeating apple's talking points as if they were gospel.

interestingly, there are never any real details about what apple actually suggested. just vague hints at an intermediate solution and an 18-month phased rollout. but of course, it is the eu's fault for not accepting that.

your gdpr argument is ridiculous. it's the user's data. if the user wants to open their data up to anthropic, openai or mistral, that is the user's decision. and if the ai company fucks up, apple would not be to blame. you are just handwaving.