btown

14 ms·

181.
by btown 7mo ago
Jokes about wallet-draining aside, we're already giving our agents a real cash budget that they use for tokens. Our harnesses have mechanisms to manage that spend. And having an easily detectable protocol would allow the harness to ens
182.
by btown 7mo ago
It's fascinating - how does one defend against an attacker or red-team who controls the CPU voltage rails with enough precision to bypass any instruction one writes? It's an entirely new class of vulnerability, as far as I can tel
183.
by btown 7mo ago
There's also just a mathematical way to look at volatility here, which is that if you look at (say) the average monthly result as a statistic for the reporting period, longer reporting periods have lower variance than shorter reporting
184.
by btown 7mo ago
I'd say that this is also true from a money-and-costs-and-value perspective. Sure, all press is good press... but any number of stakeholders would agree that "we got some mindshare by proactively protecting against an emerging thr
185.
by btown 7mo ago
IMO while the bar is high to say "it's the responsibility of the repository operator itself to guard against a certain class of attack" - I think this qualifies. The same way GitHub provides Secret Scanning [0], it should ale
186.
by btown 7mo ago
In a forum/community context, speed is vital! If it takes an order of magnitude more time to generate responses like yours and mine, one must choose which conversations one participates in much more carefully, and every such investment
187.
by btown 7mo ago
This seems way cooler than just computation (which is easy to hand off to a tool, and arguably more predictable that way). The broader point here is that you can have your model switch dynamically to/from a kind of attention that scale
188.
by btown 7mo ago
For all the challenges that AI poses to online communities, it does allow people for whom typing and dictation are painful, difficult, or impossible, to participate in those communities in ways they never could before. I think HN is broadly
189.
by btown 7mo ago
Great instincts! It would be less the entourage and more an accredited travel agency with established reputation. And absolutely correct that the skip should be auditable and intentional - and having support at the provider level for this m
190.
by btown 7mo ago
Great to see innovation in this space! If I could make one giant request, it's around giving (properly authorized) humans the ability to override the system when needed. When you make a simple API, it's all too common for a compan
191.
by btown 7mo ago
A counter-argument here: if a private company knows that its technology may be used for human-not-in-loop targeting/surveillance, and knows that its technology is not yet ready to fulfill that use case without meaningful unintended
192.
by btown 7mo ago
I don't know what sales numbers look like, but from my perspective as a casual reader who's been trying to keep up with Hugo Award nominees of recent years... science fiction may not be trendy on BookTok, but it's far from de
193.
by btown 7mo ago
They can go hand in hand! But if you give a dump of a session to someone, with literal reams of command inputs and outputs etc. interleaved in the session... they'll most likely read the beginning and the end. And possibly absorb the i
194.
by btown 7mo ago
Something I've recently come to appreciate is that Claude, with the context of your codebase and your ORM models and how they connect to your frontend, given read-only access to production databases (perhaps proxied to anonymize client
195.
by btown 7mo ago
I'm discovering new possibilities all the time with how Claude can work on a new type of task in our codebase and business more broadly. While a lot of this can be brought to the team by saying "encapsulate what you just did into
196.
by btown 7mo ago
Thanks for bringing this up, and you're right that this is closer. I still think it's imperfect, because a gig economy worker who works 35+ hours per week would be considered "employed full time" (footnotes, https:&#x2F
197.
by btown 7mo ago
In all seriousness, I’m unsure that official job numbers (even if they weren’t intentionally distorted, which is a big if these days) have caught up with the gig/creator economy. If a person making ends meet with food delivery and a fe
198.
by btown 7mo ago
> The second time the same (or similar) input is used these states are already created and it is linear. Does this imply that the DFA for a regex, as an internal cache, is mutable and persisted between inputs? Could this lead to subtle d
199.
by btown 7mo ago
> Readers are simply more willing to tolerate a lightspeed jump from belief X to belief Y if the writer himself (a) seems taken aback by it and (b) acts as if they had no say in the matter - as though the situation simply unfolded that w
200.
by btown 7mo ago
This is also just the direction that AI is taking us, even for people who wouldn't describe themselves as traditional developers. Setting aside on-device LLMs, one needs RAM and disk space just for the multiple isolated Claude Cowork e
201.
by btown 7mo ago
Some press coverage (though I highly recommend just reading the paper linked as the OP, it’s quite approachable to skim without prior knowledge, and you get to see how they turn the Star Trek replicator problem into “just” a loss optimizati
202.
by btown 8mo ago
There's also a reasonable alignment between Tailwind's original goal (if not an explicit one) of minimizing characters typed, and a goal held by subscription-model coding agents to minimize the number of generated tokens to reach
203.
by btown 8mo ago
Art and engineering are both constrained optimization problems - at their core, both involve transforming a loosely defined aesthetic desire into a repeatable methodology! And if we can call ourselves software engineers, where our day-to-da
204.
by btown 8mo ago
Compaction includes all user prompts from the most recent session verbatim, so that's likely what's happening!
205.
by btown 8mo ago
The saving grace of Claude Code skills is that when writing them yourself, you can give them frontmatter like "use when mentioning X" that makes them become relevant for very specific "shibboleths" - which you can then u
206.
by btown 8mo ago
OpenAI, frankly, benefits from the "nobody ever got fired for buying IBM" phenomenon. And there's a reason that OpenRouter has an OpenAI compatible layer highlighted not deep in docs, but on their Quickstart page: https:&#x2
207.
by btown 8mo ago
I'd ask the inverse of the question: morally, should a single gatekeeper have the right to deny two consenting parties the ability for one to run the other's software? Especially when that ability has been established practice and
208.
by btown 8mo ago
For speculative decoding, wouldn’t this be of limited use for frontier models that don’t have the same tokenizer as Llama 3.1? Or would it be so good that retokenization/bridging would be worth it?
209.
by btown 8mo ago
Do skills extracted from existing codebases cause better or worse code in that they bias the LLM towards existing bad practices? Or, can they assist in acknowledging these practices, and bias it towards actively ensuring they're fixed
210.
by btown 8mo ago
If it were in the context of parachuting into a codebase, I’d make these skills an important familiarization exercise: how are tests made, what are patterns I see frequently, what are the most important user flows. By forcing myself to dist

More ›