Josh Triplett @[email protected] @josh_triplett

Free and Open Source Software developer. #rustlang developer, working on the language, library, and Cargo. he/him Fediverse: @[email protected] github.com/joshtriplett/ Joined May 2010

Tweets

4K
Followers

4K
Following

203
Likes

3K

Christoph Heilig @ChristophHeilig

2 weeks ago

1/8 🧵 GPT-5's storytelling problems reveal a deeper AI safety issue. I've been testing its creative writing capabilities, and the results are concerning - not just for literature, but for AI development more broadly. 🚨

26 54 378 72K 226

Meredith Whittaker @mer__edith

4 weeks ago

New Sunday Times profile in which I succeed, like a fencer in a 2hr marathon match, in fending off Qs abt my personal life & consistently turning focus back to my work & ideas. Contra the interviewer's claim, many ppl do know me! They're called friends, & you know who you are ❤️…

33 115 780 55K 116

Download Image

Josh Triplett @[email protected] @josh_triplett

4 weeks ago

FAFO can be interesting to observe from a minimum safe distance, assuming that 1) there exists a minimum safe distance, and 2) the FO is happening to the people who FAed, not to other people.

0 0 3 394 0

Owain Evans @OwainEvans_UK

2 months ago

In a more practical setup for distillation, the teacher is a misaligned model and generates reasoning traces for math questions. We filter out traces that are incorrect or show misalignment. Yet the student model still becomes misaligned.

8 45 845 80K 84

Download Image

Owain Evans @OwainEvans_UK

2 months ago

New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵

292 1K 9K 1.9M 5K

Download Image

Josh Triplett @[email protected] @josh_triplett

a month ago

Python 3.14, or, πthon

2 1 12 469 0

METR @METR_Evals

2 months ago

We ran a randomized controlled trial to see how much AI coding tools speed up experienced open-source developers. The results surprised us: Developers thought they were 20% faster with AI tools, but they were actually 19% slower when they had access to AI than when they didn't.

236 1K 7K 3.6M 3K

Download Image

Rob Bensinger ⏹️ @robbensinger

3 months ago

Senior White House officials, a retired three-star general, a Nobel laureate, and others come out to say that you should probably read Eliezer Yudkowsky and Nate Soares' "If Anyone Builds It, Everyone Dies". Preorders are live.

7 47 306 60K 49

Download Image

Owain Evans @OwainEvans_UK

6 months ago

Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. This is *emergent misalignment* & we cannot fully explain it 🧵

438 983 7K 1.9M 4K

Download Image

Thorne 🌸 @ExistentialEnso

3 months ago

lmao

23 172 2K 61K 98

Download Image

Harlan Stewart @HumanHarlan

3 months ago

Palisade Research @PalisadeAI

4 months ago

https://t.co/tgvB6ZLc9x

517 2K 10K 6.0M 5K

17 121 789 136K 113

Download Image

Nate Soares ⏹️ @So8res

5 months ago

Whether or not criminals deserve due process is beside the point. The point is that due process is how the state determines whether or not someone is a criminal.

7 16 336 17K 9

Josh Triplett @[email protected] @josh_triplett

4 months ago

Sufficiently advanced incompetence is indistinguishable from malice. But for the responsible organization, the process of *solving* incompetence is very different from the process of solving malice.

3 0 2 548 0

Patrick McKenzie @patio11

4 months ago

*rereading* I try not to get self-indulgently angry often but correct amount of anger is non-zero.

2 5 63 7K 5

Download Image

Josh Triplett @[email protected] @josh_triplett

4 months ago

Disappointed in @Lenovo's support in the UK. It doesn't count as "next business day onsite" service if you first take a week to get a part. That's a week with a nearly non-functional laptop, while traveling. I expected to be able to rely on "next business day".

0 0 2 287 0

Jason Bosco @jasonbosco

6 months ago

During a call today, a prospective user evaluating Typesense asked me if we've raised VC money. I gave them my usual spiel about how/why we've chosen not to raise VC despite inbound interest, we're fully revenue-funded and happily profitable, etc. Then they tell me that the…

12 10 161 18K 34

Charlie Marsh @charliermarsh

6 months ago

We switched to zlib-rs and saw a ~10% speed-up on ARM (and zlib-ng-equivalent performance on x86, which itself is a huge feat). This also let us remove our cmake dependency 🎉 zlib-rs is so easy to build that we can use it on _all_ platforms (s390x, PowerPC, etc.).

9 25 350 21K 36

Download Image

James Miller @JimDMiller

6 months ago

"The best answer to the question, 'Will computers ever be as smart as humans?' is probably 'Yes, but only briefly'" --- Vernor Vinge. We are in the 'briefly'.