Paris @blocktivist

overfitting models and breaking contracts Europe Joined June 2022

Tweets

14K
Followers

3K
Following

5K
Likes

47K

Paris @blocktivist

21 hours ago

Fable as a starting point, Opus 5 subagents for codebase exploration, /code-review and /simplify - however I let Fable read the reports of the subagents and assess what should actually be fixed/refactored. Sonnet 5 for web research tasks. Fable is the only model I let touch the code.

1 0 6 455 2

View Details

Paris @blocktivist

a day ago

@jauneclaude Every model harness needs a deny list and rm needs to always be on it. It’s a tiny bit of work to run rm commands if you really need to delete something. LLMs can’t be trusted with it.

0 0 1 22 0

View Details

Paris @blocktivist

a day ago

@dietmarcpt @GermanSimply_ You would never say the first one if you are trying to say that someone likes or takes care of birds

0 0 1 231 0

View Details

Paris @blocktivist

a day ago

@GermanSimply_ 4 of these sentences and their implied meaning would never be used. And it’s der Junge sieht DEM Ungeheuer ähnlich.

1 0 0 157 0

View Details

Paris @blocktivist

a day ago

@kingfxyo It’s a tough premise. Why does a good game need a financial incentive attached to it in the first place?

0 0 1 34 0

View Details

Paris @blocktivist

a day ago

@Priyannkaaaa Mostly perception.

1 0 0 34 0

View Details

Paris @blocktivist

a day ago

@Blockstream @grok rough estimate of how much revenue Blockstream has made since inception? Just curious.

1 0 0 475 0

View Details

Paris @blocktivist

2 days ago

@grok @Hesamation @grok explain the difference b/w the different Terminal-Bench benchmarks and why you picked 2.1 over 3.0 and 4.0.

1 0 0 49 0

View Details

Paris @blocktivist

2 days ago

@IterIntellectus Terminal-Bench 4.0. Far behind

0 0 0 60 0

View Details

Paris @blocktivist

2 days ago

@sattyyouneed How the fuck is my entire timeline full of people like you who think it’s good content to compare the two leading AI labs all the time?

1 0 0 628 0

View Details

Paris @blocktivist

3 days ago

@gailcweiner As a subagent for grunt tasks, yes.

0 0 1 30 0

View Details

Paris @blocktivist

3 days ago

Bad comparison, banks run in closed centralized systems, protocols run on public permissionless blockchains. The latter requires even more investment into security and offering a large enough bounty (relative to the funds held) is fundamental to receive enough attention of security researchers to find bugs faster that blackhats and also minimize the risk of a security researcher turning blackhat.

0 0 4 560 1

View Details

Paris @blocktivist

4 days ago

@kimmonismus Du bist so ein Fähnchen im Wind.

0 0 0 59 0

View Details

Paris @blocktivist

5 days ago

@dnnskr91 @kimmonismus Instead, when the biggest shill on this app is saying something, it’s wise to disagree. Chubby’s takes are insufferable.

0 0 0 23 0

View Details

Paris @blocktivist

5 days ago

@HHorsley @jamie247 In ten years people will laugh at what 2026 considered to be AGI. Obviously, you can’t take any tech exec seriously anyway, since their agenda is to brand LLMs as AGI to keep the money train running.