ibejoeb

That's what I was thinking of, except it's not even dealing with the corpus of kung fu and doing the super-speed rounds of learning. It's directly reconfiguring the neurons. It's even better.

As I read it, this is saying it can effectively spot-edit the weights without training.

> Zero-data capability steering

Presumably it can de-nerf a model with onboard safeguards.

Why would any of these jobs be new? People are already doing them. Would you not just reduce the earning potential by removing tips?

Maybe I'm jaded. Is this discord announcing that it is entering the training data business?

In general, and especially for the technical roles, they are probably going to have trouble finding relatively similar compensation.

Looks like they've been around for 25 years.

Ukraine's position has been that starlink is used exclusively for drone defense, i.e., interceptors. That is sensible because that's within the contracted terms. However, there is intentional obfuscation of starlink and starshield in all of this. Ukraine also claims that it doesn't use star-anything for offensive drone attacks, which is kind of unbelievable. The contention started when the White House told Ukraine that refineries were off limits, and then Ukraine struck refineries anyway. It seems clear to me that 1) they are using the technology, however it is branded, to reach and possibly cross the border, and 2) this was a signal to abide by the temporary restrictions, as the tri-lateral talks are being negotiated.

It doesn't proxy the images, so they won't load with enhanced tracking protection enabled.

"memory leaks here aren’t too relevant" will be a real candidate for Rasmus' epitaph

context: https://news.ycombinator.com/item?id=50176735

That's not sovereignty. Sovereignty doesn't imply anything and it is not a designation bestowed by some other entity. It is purely descriptive of an evident state. If you can defend your claim to sovereignty, you are sovereign. Moreover, no claim need be made.

It's the architecture. That's the correct answer here. The current models produce good implementations. They're also quite good at identifying and planning for edge cases. But, today, a successful software project requires picking the right atoms for the job.

Some of the calculus for that picking will change, since volume of code that must be produced becomes less of an issue. And I don't doubt that models next year and year after will be able to make better formative architectural choices. But as it is now, I'm certain that actual systems handling real workloads require a human designer.

Someone who knows what good software looks like is empowered with agents. Someone without that knowledge isn't going to create a high quality system yet.

She was truly ahead of her time.

The prosecutors aren't on the hook, anyway. They have absolute immunity. The decision to charge is protected. The prosecutor would have to have done one of the few, enumerable things outside the scope of the role, like conducting an investigation without probable cause or hiding exculpatory evidence.

Not only that they can review flagged content, but they tend to have separate retention policies for flagged content. Anthropic's is this: "We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years if your chat or session is flagged by our automated trust and safety systems as violating our Usage Policy."

So don't run for office or anything like that. Someone, somewhere will have a contact that will get that.

Location: Miami, NYC, LA

Remote: Preferred over relocation, but happy to travel to office and clients

Willing to relocate: Yes, if it makes sense. USA preferred. US/EU citizen.

Technologies: ios and Android apps (Expo, Swift, Kotlin/Jetpack Compose); web applications (many: Next.js, Django, Rails, Laravel...); POSIX; sockets; .NET; LLMs (langchain, pytorch); agentic development

Résumé/CV: https://gist.github.com/ibejoeb/902026c035a995d5719204511092...

Email: joe@bokengroup.com

Well now you've got me backed into a corner...

The AI is pretty benign, too. They offer diverse models and they make it as privacy-preserving as possible, considering you can opt to use closed models. They are very open about training and retention: https://help.kagi.com/kagi/ai/llms-privacy.html. An interesting note there on models served by OpenAI: "The Assistant does not include a unique user identifier for these requests." I don't know if you can trust OpenAI, but Kagi is at least doing its part.

Also, All the AI integration into the product is gated by a single setting.

In addition to what other have said about extensions, ad blocking, no telemetry, and some nice features like tab management, it works in the Apple ecosystem. If you use a mac and an iphone, you get the native sync services for history, bookmarks, passwords, etc. Makes sense that they're focusing on that.

Developing internally initially, as opposed to making it open source, was probably the only practical way for a small team with a specific goal to get there. I don't see any problem turning it over to community development for secondary platforms. If it gets adoption, great. If it fizzles out on linux and windows, nothing is lost.

I will never love anything as much as Alec loves oil bath motors.

I think it's pretty subjective if you mean "lead" to be capability and not raw number of users. I jumped back and forth quite a bit last year because there were some pretty major shortcoming in both. Now they're both quite reliable without too much hand-holding. Codex now consistently works better for the kind of work I'm doing, and it's good enough that I'm not inclined to go to claude, because I don't have any significant problems.

I posted 2 sources. I did not post a comment. Take it up with the authors.

That's all in the cited sources. I didn't offer an opinion. I added contextual information related directly to your positions.

I don't think that's actually what the FAA has reasoned.

FAA calls all of these devices, in general, child restraint systems. Rigid seats with harnesses are allowed under FAA rules. The real consideration is blocked egress, since a large, rigid device strapped to a seat could actually make egress more challenging. That is covered in a more subjective advisory circular, which allows airlines to make policy, e.g., it may only be used at a window seat of a non-exit row.

Since FAA just this month has decided to start enforcing a 1992 rule requiring even so much as a paper cup to be stowed before pushback citing the egress concern, we might even expect some new policy on this.

For context:

Illegal immigrant voted: https://www.justice.gov/usao-mdfl/pr/colombian-national-plea...

6,600 noncitizens registered, 400 noncitizens voted, undisclosed how many, if any, are unlawfully in country: https://www.ap.org/news-highlights/elections/2026/6600-nonci...

I use kagi exclusively because it works. I get the results I need every time, and it's not annoying. I haven't used Google in years. Simply no need. I really hope people keep paying kagi so they stick around.

Thoughts and open (I think) questions:

1. What are we actually talking about, slowing development or slowing public release? I can't imagine the the US is keen to slow actual development when we have adversarial state-backed labs continuing development. Seems to me that this is a global pact or it's nothing. Stopping a single lab, even if it's the current leader, won't put the genie back in the bottle.

2. Unless they stop actual advancement, isn't the risk essentially identical? At this point, it's an arms race between the lab and its own model. I don't think it far fetched to think that it might successfully exfiltrate its weights, for example.

3. There are going to be factions who don't want to stop. Should we worry about them defecting to other labs or other countries?

Then released under the MPL

I don't know... WordPress is a pretty huge open source project and community. He also opened Pocket Casts.

The PM got his rating about 18 months ago. The PF only got his this past May. The PM had more overall hours, but this is looking a lot like a skills issue. He was flying an unstable approach. He wasn't configured for landing.

This pairing, considering past incidents with similar airframe vs. total hours disparity, really does look to be bad practice. You've got to acknowledge the human psychology. Part of it is probably the crazy pipeline over the past few years where people are racing to 1500 hours no matter how they log them, or they're getting on restricted certs with lower hours.