Leynos
Leynos on Hacker News
Ask the model to declare its assumed axioms. Agree that those can be trusted for the time being. Things like the behaviour of third party libraries. Ensure that your review agent enforces this.
https://www.lemonde.fr/en/pixels/article/2026/09/26/openai-s...The users in this case had given permission for their data to be used for training (or at least hadn't opted out). They had not, of course, given permission for their data to be shared publicly or with a third party in this way. Consequently, OpenAI categorized this as a breach (i.e., something they did not intend to do), and sought to rectify the situation.
That's what discovery is for.
Did you opt out of data retention?
"AI skills" are the same skills we always needed:
- Learning in a self-directed way that aligns with the needs of your role and where it is going
- Clear communication, the ability to edit and know how to say what is important
- Forming and testing hypothesis
- Evaluating and testing tools and work product, and being able to recognize and rectify deviations
- Delegating, setting constraints, rubrics and escalation criteria
- Having a clear definition of done and being able to stick with it
- Understanding business domains and modelling them usefully
- Collaborating with others, communicating your needs and capabilities clearly, eliciting the needs and capabilities of others, synthesizing conflicting needs and aligning compounding capabilities
The adoption of genAI and agentic tooling really just serves to make all of these more important.
They're talking nonsense. Any actual research lab would have zero data retention enforced.
OpenAI and Anthropic wouldn't be able to get enterprise contracts without it.
Sure, the AI vendors could be lying, but I think it unlikely given the harm it would do to them to violate an enterprise contract in this way.
The Bridge trilogy was the Gibson I grew up with. The Sprawl books felt almost like archaeology. Blue Ant was where he really hit his stride for me.
Because it is not a prng. If you want a prng, don't use the thing that is not a prng (or at the very least, ask that thing to use one)
Nice. I missed that. This is good news.
So that's three now. This, tsgo-rs, and `bun check`
4.1 Flash on medium is a solid builder agent
It isn't just restricted to the exploits it is programmed with (and without human intervention to load additional exploits).
Of course, it remains to be seen what several thousand 27B models running at sub 1 tps speed can achieve. (The IoT device mechanism proposed by Anthropic).
I suspect a more likely path is compromise of the billing and provisioning systems of fly-by-night neoclouds.
When I was learning to drive, I was taught to stay within the speed limit.
Slack is also allowance in budgets (time, finance, materials) to allow for unplanned pressures, demands or constraints.
The model was sharded into ~800MB pieces over devices with 4GB RAM. Typical device storage is 16-32GB NAND. Throughput was 0.39tps.
Compromise enough devices (think thousands of off-brand TVs), and you have plenty of redundancy.
More likely the author squashed the commit history.
You start with least privilege. A shared instance of an artifact host is very far from least privilege (I'm comfortable in accepting that OpenAI did not wilfully create this situation, but they were certainly negligent).
And yes, it does indeed seem that a UK government agency ran a controlled experiment where they allowed an agent to instigate phishing attacks against members of the public. Attributing this to Anthropic is wrong. I'm not sure how I feel about my government conducting such attacks on non-consenting individuals, but I'm not comfortable about it in the slightest.
I did not say this shit is easy.
I think that you're arguing against something different. Visiting your MP at their surgery or lobbying them at their office in the house of commons is not the same thing as buying them two and a half grands worth of spectacle frames.
> Any actor that's at all serious in their field is, sooner or later, going to hit a point where LLMs cease being useful to them and actively hinder their further progress, and we absolutely do not see the massively overwhelming advantages that LLMs would need to have in order to become inevitable.
I can only conclude from this that I am not at all serious in my field. That's okay. I'm too busy enjoying myself.
Inadequate or broken safeguards do not fall under my definition of reasonable.
The precedent I would point to are Dangerous Dog Laws. We hold the owner responsible for the behaviour of their dogs.
Revisit later if the comparison starts to become ridiculous.
The scenarios set out in the article happened. And it is generally accepted that negligent behaviour should lead to consequences.
Google yet again showing their leadership in product delivery, marketing strategy and bureaucracy
HackerNews does love a good conspiracy theory.
This is an old trick in the prose generation community to elicit more honest critique. See the "drunk claude" and "stoner buddy" prompts.
I'd start with something research focused like Undermind or Elicit. Although I don't think that the author is comfortable with using a tool that isn't produced by the model lab.
The planning model for tool use sounds something like CaMeL, which someone should really try implementing in a product.
Does dropping a prng in place of Jev produce similar results?