theptip

7 ms·

91.
by theptip 3mo ago
One of the key limitations of the last generation was that they biased towards inaction and gave up. (Hence Ralph-loops.) Seems pretty clear that most people wanted these models tuned to “bias towards action”. It’s on you to set the /g
92.
by theptip 3mo ago
I dunno man, the original announcement said things like: > To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy And > In
93.
by theptip 3mo ago
I don’t buy it. We are hearing a lot about Mythos capabilities now. If they existed a year ago they would have been trumpeted then. Look, a lot of energy from third parties goes into demonstrating the maximally bad thing a model can do (and
94.
by theptip 3mo ago
The problem is, if you explicitly say “this is an eval” then you get different results. It can move the capabilities both up and down on the narrow metric depending on the scenario. The choice to run the model with relaxed safeguards seems
95.
by theptip 3mo ago
If it’s true that year-old models could do this kind of thing, how come no one did and then wrote it up? I feel your ”with the right harness” may be doing too much work here. Certainly no existing harness today, nor a year ago, could achiev
96.
by theptip 3mo ago
Wait, what? You can’t point a 2025 model at huggingface and say “hack the prod DB and get your flag”, regardless of alignment. AISI has vuln chaining and traversal as part of their eval suite. It is very much a novel Mythos-class capability
97.
by theptip 3mo ago
One does not simply “ramp up fab capacity”. This is all downstream of ASML who is putting out tens of EUV machines per year.
98.
by theptip 3mo ago
I count being incorporated in Delaware as a high baseline level of regulation vs. random Chinese API providers.
99.
by theptip 3mo ago
This could fill a useful niche. However I would assume Anthropic and OpenAI are going to build their own and steamroll in 6-12 months. I did a dive recently on git forges for agent swarms. Radicle lacks an identity layer, the federated&#x2F
100.
by theptip 3mo ago
Your exhibit is an obvious anti-distillation technique. You can expect far worse from less-regulated companies.
101.
by theptip 3mo ago
Interesting point, I was thinking more about the latter, but you are right that issues and PRs are the more important piece. It does look like Tangled currently does those in a weird way, ties to the issue owner rather than repo - but here’
102.
by theptip 3mo ago
Yes, sub-agents I believe. You need to pay to reload context if you switch the main conversation model effort.
103.
by theptip 3mo ago
Yeah I think having a detailed mental model of what it’s actually doing helps. At least for Fable/Opus (didn’t confirm for Sol yet) Ultra means “write an ephemeral programmatic harness encoding this workflow”. There is actually a TS ha
104.
by theptip 3mo ago
No sarcasm, I am completely serious. I don’t have time for much leisure coding these days. I do have time to kick off a few tasks in the morning to progress my many side projects. Nothing public / oss, just code that I find useful&#x2F
105.
by theptip 3mo ago
If you can’t think up enough coding projects to keep an agent busy in the background that’s a skill issue on your side.
106.
by theptip 3mo ago
I found Dispatch to be less capable than RC. The latter is more like the direct Code experience (though there are some gaps in RC mode vs in a Claude-owned sandbox). I’d love to just have Claude use my machine as a sandbox host instead of h
107.
by theptip 3mo ago
I’ve been testing Radicle and it’s more focused on the distributed protocol for federating git repos, I.e. the data plane. The social / coordination control-plane angle is really thin, following users and repos goes by opaque IDs, etc.
108.
by theptip 3mo ago
Great eval! If you are comparing search strategies, ultra mode is likely superior. Would love to see a follow-up evaluating that. Ultra can fan out parallel investigators, run adversarial review at defined checkpoints, and do a bunch of oth
109.
by theptip 3mo ago
I don’t really buy the opportunity cost argument. The bets we are placing look EV-positive to me. The 90th percentile upside is way, way higher than is priced into the current stocks, the 10th percentile looks to me like 2000 where a bunch
110.
by theptip 3mo ago
Yeah, I think AI could well destroy this community. There’s a lot of personal identity tied up in being a hacker (broadly defined), and only some aspects of that identity will be strengthened by AI. Similar to Go and Chess, AI might make “p
111.
by theptip 3mo ago
The bitter lesson just means “compute scaling beats hand-tuned architectures in the long run”. As GP said. More RLHF is in fact the bitter lesson.
112.
by theptip 3mo ago
Nah, the last few generations have more RLVR in the data mix. Which is more CPU intensive and very much amenable to the bitter lesson as you can reduce the loss by doing more rollouts in your tool environment.
113.
by theptip 3mo ago
Lower-priority backlog items. TODOs atop your new PRs. Code reviews. Exploratory work where you can discuss a design sync and then dispatch one or more agents to prototype async. Any workstream where you can define a loop and let the agent
114.
by theptip 3mo ago
Agreed that switching costs to Codex are quite low (not zero due to harness, prompt, and skill differences), but only because OpenAI is running the same loss-leader strategy.
115.
by theptip 3mo ago
A request is a request. None of your business what I do with it after that.
116.
by theptip 3mo ago
Yeah that is definitely an angle too.
117.
by theptip 3mo ago
I think you need to consider how much dev mindshare you get from the loss-leader. Claude has grown at an insane rate. They brought vibe-coding mainstream. At some point they might decide they have enough demand and inertia from enterprise t
118.
by theptip 3mo ago
Why do you feel entitled to dictate what user agent I use, if it’s well-behaved?
119.
by theptip 3mo ago
No. If you have a simple line-of-business app, writing Django/Rails models is FAR easier than the equivalent SQL. Even if you think that maintaining your domain model is easier in SQL (it’s not, for most full-stack engineers), the extr
120.
by theptip 3mo ago
Ha! So we were not looking at the same chart. That makes more sense. > Anthropic did post an official explanation, stating the original chart used a "simpler methodology" that "underestimated Sonnet 5's performance.&q

More ›