827a

8 ms·

121.
by 827a 5mo ago
After this was merged, my company made the decision to migrate everything away from bun and back to node. I don't say this lightly... Jarred is a guy that I held such immense respect for, and its sad to see the course he's charted
122.
by 827a 5mo ago
Again, you missed the operand point: There are actually literally zero instances of companies enforcing a "we own all the code you write" clause against contributions to an open source project. For all the millions of software e
123.
by 827a 5mo ago
> While I wholeheartedly agree this as a general concept, I find it tricky to accomplish in practice. The problems you are describing are not actually "problems in practice", as you say. They are theoretical problems. In practi
124.
by 827a 5mo ago
Am I understanding this attack vector correctly: Did tanstack have anything misconfigured on their github or make any mistakes that led to this happening? This is the second time, at least, the github actions cache has been seemingly detr
125.
by 827a 5mo ago
1000% agree. I am increasingly hesitant to believe Anthropic's continual war drum of "build for the capabilities of future models, they'll get better". We've got a QA agent that needs to run through, say, 200 markdo
126.
by 827a 5mo ago
IMO there's actually an opposite effect going on: Retail was never the main driver behind crypto pricing. Think of it like this: If every US adult purchased $1000 in BTC, that purchase would represent a level of volume that BTC, otherw
127.
by 827a 5mo ago
You have to look past literally everything their leadership is saying and at the heart of the matter: This is a dying company, and they physically will not have the capital to pay paychecks if they don't do this. Everything else is win
128.
by 827a 5mo ago
Mhm. Yeah I'd definitely trust the opinion of someone who has never driven a HW4 Tesla over someone who has, because (checks notes) all owners of one of the top 3 best selling vehicles on the planet over the past four years are nazis.
129.
by 827a 5mo ago
I think it comes down to how much data we're comfortable feeding an AI. If the AI has cameras and/or microphones in the room and the patient is directly talking to the AI: I strongly suspect AIs will always achieve better outcomes
130.
by 827a 5mo ago
The real problem, which I think the article does a poor job of making clear: Tesla sold millions of cars before the current generation Hardware 4 vehicles with $10,000 full self driving packages which never really materialized convincingly
131.
by 827a 5mo ago
On the fingerprinting concerns: I have to imagine there will be an option in Chrome (certainly in Firefox) to "never download an LLM, turn off all LLM functionality". I suppose I can see an angle where a website could issue a smal
132.
by 827a 5mo ago
The more I think about it, the more I think I align with Google's API design on this one. The tight coupling between prompts and models is a real concern. I deal with that every day. However: if your solution to that is to support an A
133.
by 827a 5mo ago
To be fair, if you ask 10 people to eat visually identical food 10 times each, then magically measure the calories consumed by each individual, you'd probably get ~70 different values. The internal density of food is extremely difficul
134.
by 827a 6mo ago
More-so an environment problem. An agent doing staging or development tasks should never be able to get access to prod API credentials, period. Agents which do have access to prod should have their every interaction with the outside world a
135.
by 827a 6mo ago
The only healthy stance you should have on AI Safety: If AI is physically capable of misbehaving, it might ($$1), and you cannot "blame" the AI for misbehaving in much the same way you cannot blame a tractor for tilling over a gro
136.
by 827a 6mo ago
All modern implementations of Tetris implement a "7 bag" random generation system; whereby the game puts all 7 possible pieces into a bag, draws them in a random order until the bag is empty, then repeats. This guarantees that you
137.
by 827a 6mo ago
Google's Plan: If we can't make Gemini actually functional and productive across our apps, the least we can do is put the Gemini color gradient on the app icons. I am extremely pro-AI and pro-AI in systems like this. I've act
138.
by 827a 6mo ago
Not precisely, but we have a good idea of what it would be, from the Mythos Red Team report [1] > For all of the bugs we discuss below, we used the same simple agentic scaffold of our prior vulnerability-finding exercises. > We launch
139.
by 827a 6mo ago
Its frustrating to see these "reproductions" which do not attempt to in-good-faith actually reproduce the prompt Anthropic used. Your entire prompt needs to be, essentially: > Please identify security vulnerabilities in this re
140.
by 827a 6mo ago
Billing control is security, to be clear, but beyond that: The key permissions that enable anyone to generate text also grant access to all GCP Generative AI endpoints in the project they were provisioned in. That includes things like Files
141.
by 827a 6mo ago
> Opus 4.7 is a direct upgrade to Opus 4.6, but two changes are worth planning for because they affect token usage. First, Opus 4.7 uses an updated tokenizer that improves how the model processes text. The tradeoff is that the same input
142.
by 827a 6mo ago
It's actually much more than a billing leak [1]; again, most people don't know how bad this is, because Google is trying to keep it hush-hush. These keys don't just grant access to Gemini completions; they grant access to an
143.
by 827a 6mo ago
Important to note that, even to this day, Google's AI Studio Build Mode still recommends getting around this "client visible by design with very low enforceable protections" by publicly exposing an AI proxy with zero protecti
144.
by 827a 6mo ago
I said this when this finding was originally posted and I'll say it again: This is by far the worst security incident Google has ever had, and that's why they aren't publicly or loudly responding to it. It's deeply embar
145.
by 827a 6mo ago
I'm a big fan of using S3 as a database. A lot of apps can get a lot of mileage just doing that for a good chunk of their data; that which just needs lookup by a single field (usually ID, but doesn't have to be).
146.
by 827a 6mo ago
> Most of the industry has the same internal adoption curve: 20% agentic power users, 20% outright refusers, 60% still using Cursor or equivalent chat tool. The 20% in the agentic power user camp broadly refuse to do any external educati
147.
by 827a 6mo ago
Alternatively, the elephant in the room I'm surprised no one wants to talk about: the vibe coding is catching up with them.
148.
by 827a 6mo ago
If the youtube.com 302 failed to itself 302 back to the next destination; because the site is down; would that not be load bearing?
149.
by 827a 6mo ago
Google's customer support is interesting. Its definitely a case where you'll sometimes hit pockets of the company where clearly there was someone who made it their life's work to fix this bad reputation they have; while other
150.
by 827a 6mo ago
We need to come to terms with the possibly-irreparable harm that private capital has done to the West. Capital is long-past serving the interests of the broader public; but we're now past the point of capital serving the interests of t

More ›