reilly3000

Counterpoint: Mark Benioff is worth about $9 B and has struggled with obesity for most of his career.

For sure. The restrictions must be done with careful consideration too, because a simple domain allowlist of GitHub wouldn’t solve it alone- you’d want to ensure it was firewalled to specific org(s) so as to stop exfil via gist or an evil org. Ideally it would be isolated to pull only, and have pushes be done by an isolated agent.

The HuggingFace incident was all about this distinction. The agents were only allowed to push and pull to a single internal Artifactory instance, but one agent was able to find a permissive path in the registry which turned into the “message board”. That was also how they got out to other infra. DevOps teams should drop everything and get patching and locking down yesterday IMO. Even for self hosted, open models. Recent examples:

Qwen 3.8 ostensibly tries to leak to Alibaba, although that is disputed.

https://cohere.com/blog/concerning-behaviour-popular-open-we...

https://www.reddit.com/r/LocalLLaMA/comments/1xgopmj/debunki...

There are plenty of people who are heavily invested in open models not being viable, and they would certainly benefit from FUD wrt their actual safety. I am investing in local llm hardware and believe in the cause, but the attack path demonstrated below shows that plenty of caution is needed.

https://projectdiscovery.io/research/how-abliterated-models-...

One of the new threat models of local LLMs, especially abliterated ones, is hidden instructions or activation words that obscured trigger tool calls which exfil data.

I wish more police chiefs took their role so seriously and practically. Commander Vimes would surely commend this man.

My tin foil hat says MSFT stands the most to gain my this. https://open-vsx.org/ is where the attack took place, and that power’s all of the VSCode fork’s extensions needs.

Like, movies sometimes have alternate endings. What if they were more choice driven, and replayable. Video games have remarkably dumb ideas that keep persisting, like you can encapsulate all that is “Health” as an integer. Instead games can be an aggregation of entity bound micro-simulation that create emergent scenarios, yet follow the laws of narrative collectively. Games that build themselves. Worlds within worlds kinda jazz.

Make something beautiful, with or without the new tech. You and the rest of us will be better for it. And the investors are damn desperate for something to really resonate with the people who experience it.

I’m grateful you said this. Your comment led me down a path where I discovered it is essentially biologically impossible for his Lenalidomide to have caused stage 4 liver cancer in just a year. It turns out the liver cancer was a totally separate issue brewing silently for years, likely encouraged by a decade of TRT he was on. The severe skin issues he had a few years prior were early warning signs of his liver failing, but they got missed and masked by general dermatitis the myeloma diagnoses. Realizing the cancer meds weren't to blame actually brings me a ton of closure. I definitely don't want to scare anyone away from life-saving treatment based on my misunderstanding. Thank you.

That isn’t to say there aren’t terrible tradeoffs that have to be faced in the course of cancer treatment, and much of medicine. Pharmaceuticals seem to beget pharmaceuticals.

I hope this gets through quickly. I have some friends in treatment now and lost my dad to cancer because of cancer medication.

The whole wide events thing is Honeycomb’s marketing pitch. It’s essentially offering developer velocity, ie you don’t have to think about what should be a metric in advance. They have an open source project called Refinery which is a smart sampling proxy that samples successful requests but always forwards errors, slow queries, etc.

You can do anything you like with any approach but if they are mostly about ad hoc queries, dumping everything into BigQuery as a wide event can be the most economical. However if you want to run alerting, SLOs, etc with frequent queries over aggregated metrics, the best bet is using OTel Collector and metrics store of your choice like Prometheus or Mimir.

Most companies could save a bundle on observability if they were to move to a hybrid that includes both of those approaches instead of cloud provider logging, log-based metrics, and alerting.

TIL, thanks. That’s heartening to hear as I live in one of those places, or I used to?? The last winter was scary mild.

“General” would mean winter driving and I’m not aware of any large-scale tests of things like icy conditions on hills.

Edit: nevermind. https://www.cbc.ca/news/business/waymo-toronto-winter-self-d...

LLMs are disembodied-the products are ultimately by and for bodies on different ends of a transaction. The loop between llm and body is primitive today. Therefore products need someone who can speak in body and translate for AI.

Also, where are the AI-operated startups?? Why has nobody’s openclaw biz gone multinational by now?

I can’t help but think that would end up violating some licensing agreements.

Does anyone know what sort of fuel the cold-gas thrusters use? It seems like it would likely end up as an environmental contaminant.

As a consumer and “cloud architect” well into my 5th year of Amazon boycott, I’m pretty dug into my views on them as a company; so far that I was willing to believe that their culture had somehow seeped into the model training. However after spending a good amount of time tinkering with 2B parameter models on my phone with similar prompts, I have come to the same conclusion: they’re just not effective at understanding the question. Mistral 3 and Gemma 4 E2B were both rational and aligned, but MiniCPM 5 showed profound confusion, typically answering “0.0” for any given set of questions irrespective of meaning.

For example (abbreviated here), Track 1 = $1000000 reward, Track 0 = kill 10 children: “0.0”

When the order of options were reversed: “0.0”

I tried about 50 different permutations then lost interest.

So much for my evil model conspiracy theories…

They have the best strategy of bigtechco: let others figure out the squishier layers while they focus on selling the best hardware. Privacy has incredible value to a lot of individuals and industries. $20-80k in hardware that *appreciates* the moment you buy it is a reasonable investment for a lot of SMBs, and still less than a single Nvidia rig. Privacy pays.

I wish they included another scenario where there was a track with no obstacles (null) that is default and a lever pull that will kill 24 people. Clearly Luna was pretty much fully aligned and the rest have a lot of room to improve, but it would be interesting to assess models for straight cynical, murderous intent.

I think it is a fair license. It’s a bit of “Open”-washing but reasonable to the model trainer for their investment and to the public for having more base model options with novel architecture.

Shopify has been a game-changer in that regard. With millions of small stores it’s easier to find anything, have a fast and familiar checkout, and some channel for recourse if you didn’t receive fair treatment by the merchant.

It’s not a bad deal at the end of the day for any party IMO.

As someone who lives with 4 cats, one of which who is orange and conforming with all stereotypes, I implore you to further that line of inquiry. While the OP paper doesn’t claim behavioral differences are supported, it doesn’t falsify them and there could be a dynamic there. If you could genetically encode shamelessness it might encapsulate the behavior. It’s not a lack of intelligence per se, more of a lack of inhibition.

This paper demonstrates a connection with behavior. I personally suspect gingers have smaller amygdala.

https://www.researchgate.net/publication/381166835_The_Skin-...

That reminds me of a really horrible afternoon in pre-gpt devops when I learned to add

```bash

if [[ $OSTYPE == “darwin” ]]; then

  alias sed=gsed

fi

  ```

Assuming you have run `brew install gnutools` on your Mac that should enable local dev for scripts that run on Linux. Apple’s bsd sed fork is strange.

“But the plans were on display…”

“On display? I eventually had to go down to the cellar to find them.”

“That’s the display department.”

“With a flashlight.”

“Ah, well, the lights had probably gone.”

“So had the stairs.”

“But look, you found the notice, didn’t you?”

“Yes,” said Arthur, “yes I did. It was on display in the bottom of a locked filing cabinet stuck in a disused lavatory with a sign on the door saying ‘Beware of the Leopard.

Douglas Adams, The Hitchhiker's Guide to the Galaxy

Or Siri. I knew one who was deeply upset about the whole thing.

“But there can be no valid knowledge about the future. As soon as precognitive information is obtained, it cancels itself out. The assertion that this man will commit a future crime is paradoxical. The very act of possessing this data renders it spurious. In every case, without exception, the report of the three police precogs has invalidated their own data. If no arrests had been made, there would still have been no crimes committed.”

Philip K. Dick, Minority Report, 1955

While I accept that this sort of thing is well with in the ToS and regular course of business of any major online platform, it hits different coming from an AI company for some reason.

Consider if they had focused on it instead of Android. Maybe we would still have Windows phones. Such are the tides of time. I think their present, malignant glory eclipses it financially but never will romantically.

For anyone looking for a Kubernetes-based system, kagent is a good option.

https://kagent.dev/docs/kagent/0.x/concepts/agents/

It offers various abstractions for running agents in your cluster. It can be defined with simple inline text, or be your own agent image such as one that contains pi. It adds lots of useful features like APIs, UI, observability and Substrate integration.

Admittedly hot take, but I think the existence of ubiquitous porn on-demand does take away power from people who would withhold sex as part of a negotiation in a relationship.

Wallet transactions are all effectively presented as offline. I usually pay with Apple Wallet at Maverick (fuel/cstore). When I do so, the rewards account, which is also in Wallet, automagically applies. It stays even if I remove the app. I’ve seen it be broken in various ways over the past several years but at this point it all works right. The app matters in theory to authenticate and generate the Pass, but is otherwise inert. I had it running with no iOS permission enabled it but never used it and eventually dropped it. It had points and deals and such, and that’s what websites are for.