YuriNiyazov

That's because "cd $HOME" is unnecessary, you can just type in "cd" and that puts you in ~/

And "$HOM" and other things like that end up resolving to blank, so "cd $HOM" just gets resolved to "cd", which, as I said, puts you in ~/

Such a misleading title. Change it to "qualified security researchers found a vulnerability in OpenAI via an official bug bounty process" and all of a sudden it is completely uninteresting. "System used as designed."

Very interesting, thank you!

Wefunder

My GH repo: https://github.com/yn/claude-output-styles

My LI post: https://www.linkedin.com/feed/update/urn:li:activity:7495167...

It's easiest to explain this while anthropomorphizing the model, I know some folks here hate that, sorry about that. I heard an interesting diagnosis for why Claude does this: the output is a compressed version of its thought traces, very dense because the model is under pressure to use as few tokens as it can and to pack as much (for accuracy) of its concepts into the output.

One of the reasons that "don't do X" type of instructions work reliably is because you are telling the model "don't think of a pink elephant". There's also Anthropic's related research that shows that when you tell a model "don't do X", and it does X later for whatever reason, it starts acting more misaligned. This is because it thinks "well, I guess I am the sort of model that disobeys instructions, whatever" - this was specifically about cheating on tests, but you can imagine this happens in other contexts as well like following instructions on what kinds of text to output.

So, what you want to do is to avoid telling Claude "don't do X", and tell Claude "in your thoughts, in memories and various notes that you write, use your Claude-ese. In your output to humans, translate everything into long full sentences."

If anyone's interested, I can share my Claude Code output style that reflects this.

(Hi Adnan! Long time! (Adnan is an ex-coworker))

Theres Sofia (ex-Uzbek) and uzbegim, what’s the third one?

The writing style is grating, it's not a pleasant read.

What model are you running? With what settings?

Don't know about GCP, but our AE on AWS was also continuously rotating, and as best I can tell, their job was to figure out what we are planning to build, and to ensure that we should always use <INSERT AWS SERVICE DU JOUR> for that, rather than a competitor product or build it ourselves.

Me too. It's great that my employer pays for it and there's basically no budget, because this configuration is 10x more expensive than the regular default Sonnet.

I got 99/100

Editing Jupyter notebooks in VSCode side by side with Claude Code extension is a pretty good workflow.

The example itself that Simon puts up is questionable. I might be wrong about this, but I thought I read elsewhere that the “buy, scan, destroy” method was explicitly not the problem, and instead the issue was that anthropic downloaded libgen, and the settlement was for libgen.

Edit: https://news.ycombinator.com/item?id=45143392

Why are you at azure and not at a frontier ai lab?

Thanks for your reply; I can’t edit the original comment but I have updated my personal understanding of Marcus’ position.

Thanks for your reply; I can’t edit the original comment but I have updated my personal understanding of Marcus’ position.

Thanks for your reply; I can’t edit the original comment but I have updated my personal understanding of Marcus’ position.

motte, meet bailey. Gary Marcus' shtick the entire time has been "LLMs are the wrong approach", and now the claim is "actually, the entire time I've been claiming something much weaker: LLMs that call out to code interpreters are sufficient for neurosymbolic AI"/

A long long time ago, at ClojureConj 2014, I asked Rich Hickey whether a cpp-based clojure was possible, and his answer was "well, the primary impediment there is a lack of a garbage collector". There were a lot of conversations going on at the same time, so I didn't get an opportunity to "delve" into it, but:

1. does that objection make sense? 2. How does jank approach that hurdle.

Ex-iron curtain, amirite?

Visibility is essential, and visibility is owned by the devops team. ALl the resources have to be tagged with the team that owns those resources, and you should have a meeting once a month or so with that team where you bring to them a report of "here are all the resources we have assigned to you, do you agree?" and if they agree, then great, they are responsible for it, and if they don't agree, then now you have to have a conversation where you bring in the PMs and the EMs, and the head of finance if you need to.

And it's essential that you check in regularly, rather than just give them a dashboard and say "ok, you go look here and tell me if anything is amiss", because they will never look.

Me too (Oracle Park), and I also relate to the not feeling old for a couple of hours - except the few times when I saw parents bring their kids to this show, and the kids were definitely not older than like, nine or ten, and were (!) mouthing the words to songs from Dookie.

First read "The Phoenix Project".

Then tell your manager to read "The Phoenix Project". Or just extract the relevant lessons directly from there and send them to him.

How does this work again?

"Sign here that you are resigning"

"No thanks, you'll have to fire me if that's what you are trying to do"

Read https://situational-awareness.ai/

If you can't develop a counter-argument to that, there's no reason to believe another AI winter is coming.

sure. email in bio.

Sounds like a company where they don’t like PIPping or firing people.

I’ve been in your situation, and IMHO there’s no amount of time that will fix it. He’s reached his ceiling.

Do the hard thing. Or transfer him to a yet another team.

The "Beware the Alan Turing Fetish" link doesn't point to an existing page anymore.

Academia.edu | https://academia.edu | San Francisco, CA | Full time | Hybrid, Onsite 2x-3x per week.

We are Academia.edu. Our mission is to accelerate and democratize the world's research because scientific discovery is based on it, and scientific discovery benefits everyone.

With over 180M free registered users and nearly half of the world’s academics using our platform every month, we are in a unique position to disrupt the $25B academic publishing industry. Our users range from researchers, academics, industry professionals and college students, with over 300,000 paying premium subscribers, contributing to a doubling in revenue over the last four years.

Our Stack:

- Ruby on Rails, Sidekiq

- PostgreSQL, Redis, Elasticsearch

- React + Typescript

- RSpec, Chromatic, Jest, Storybook

- CircleCI, Jenkins, Ansible, Terraform

- AWS ecosystem (EC2, S3, RDS, Redshift, many others)

- You don’t need to know any or all of these, but must be willing to learn!

What you’ll work on

You'll be designing and building systems to make our user base more productive on our platform. Here are some things that we are currently working on:

- A fast and scalable peer review system that identifies and distributes high-quality academic works

- Services that help academic authors understand the impact of their scholarship via text parsing and computation of bibliometrics

- Algorithmic approaches to identification of academic content and disambiguation of authorships in order to make every paper ever written available for free

- A paper recommendation engine that helps users discover the most relevant research

- Search Engine Optimizations to improve the discoverability of our corpus of 50M+ PDFs

Compensation: Base Salary + Equity. Base range: $160,000 - $240,000

Apply at https://jobs.lever.co/academia/ee3c7f6d-ec42-46ca-ad65-53846... and also email me at yuri@academia.edu (I'm the VPE) to tell me you applied so I can fast track you.