nsagent

It's for my startup. I wouldn't have bought it otherwise.

The fact that I can get this in my hands up to two months earlier than the 256GB version (Nov 25-Dec 3 vs Jan 15-29) is a major factor.

Bought it without looking too closely to make sure I got in before the rush (comes Nov 25-Dec 3 before my M5 Ultra 256GB which is supposedly shipping end of January).

After taking a look the $11.5k premium over the 256GB might not be worth it. Gonna have to mull over the decision and possibly buy two 256GB machines instead (which might be the smarter option).

I interned there for a summer working on evaluating text generation and despite the fact that I used no consumer data whatsoever, but rather a well-known public research dataset, I also couldn't egress from the machines I ran training on. Not even reports on the training (admittedly, this was 2021 when they had just introduced this new system). So it was even more of a damper on research progress than you might imagine.

Why recompile the harness? Why not just allow a flexible plugin system like maki [1], which takes its plugin philosophy from Neovim.

[1]: https://maki.sh/#lua

While AirBnB might be bad, my mom booked a place using Agoda. It was a NIGHTMARE. Admittedly this was in Egypt, but the unit was filthy and seemed to be owned by a company in the UAE.

If it was just the filth, it could be like any other terrible experience, but this mid-rise was standing next to the rubble of a razed building beside it and there was literally an open gas pipe hanging upright two floors down with people smoking above. The doorman who showed us the unit seemed unconcerned and it was nearly impossible to get our money back until we literally threatened to sue Agoda. We had sent them all our evidence (we took pictures) and even had the doorman talk to them on the phone and it was still like pulling teeth to get them to refund us.

To add insult to injury the place was more expensive than nearby hotels and I just checked and the place is still listed (we went a year ago)!

Gotta love the disclaimer at the bottom:

  Microsoft remains OpenAI’s primary cloud partner.

Wow, nearly every top level comment is about coding their own version of this.

I'd love some perspective on what problem this setup solves and how do you address sandboxing.

This is definitely the most defensible position. I'm sure others want stronger protections, but taking lawsuits off the table for any item no longer for sale (be it a game, other software, hardware, media, etc) would make a ton of sense.

No one can reasonably claim to be harmed in such a scenario. Enshrine abandonware into law.

I remember the same doom cycle during my PhD when GPT 4 came out [1,2]. It seemed like everyone in NLP was freaking out except me. A few years later and new CS PhDs are still graduating and joining academia and industry. Much ado about nothing.

Considering I currently live in Ithaca and have a friend doing a Math PhD, so I've heard from him some of the discussions in the department. I think the biggest shock for people is that individuals spent significant portions of their time on some of the problems that OpenAI supposedly solved (I say supposedly because all of the proofs have yet to be independently vetted). I understand that it's a shock, but I'm confident that in time it will be no more of a blip than it was for NLP.

[1]: https://xxcancel.com/mayhewsw/status/1636476184098643968

[2]: https://xxcancel.com/andriy_mulyar/status/163613925780582809...

He was a lead engineer for their AI agent platform.

https://www.linkedin.com/in/alfozan

Dreams did use a standard ECS [1] and despite being a first-party title it's not widely considered to be a AAA game (based on size of the team and budget compared to typical AAA games). There might be some AAA studio that eschews the approach, but it's very common for speed and efficiency, especially on consoles.

That said, I'm glad you shared your thoughts and experience. It's helpful to see what people try.

[1]: https://youtu.be/mFBmoCv5EcQ?t=1982

  Future work will extend Minions to support distributed model training in addition to inference

That could allow enthusiasts to pool compute resources on devices like Mac Studios, Mac Minis, etc to perform large scale federated training. While it'll be massively inefficient compared to running in a data center, it could democratize open weights models even more.

It would be especially useful for RL post training of existing models, which is where much of the agentic abilities arise from.

Reminds me of Incredibuild.

The sun's radiation causes cancer, yet placing a blanket warning stating the sun causes cancer likely leads many to wholesale avoid the sun, which would be folly (humans need some amount of sunlight).

Similarly, imagine if nutrition facts labels only had to state whether the food could be potentially unhealthy, without any other nuance.

A sensible version of Prop 65 would acknowledge the specific substance, its dosage, and the best known safe dosage. Equipped with that information a consumer can make an informed decision.

The this is much ado about nothing. Every single entry on the page you linked goes to its product page with the exact Unix standard it is certified as. Once you select a filter, you cannot get back to unfiltered list without the back button on the browser.

This is just a site rendering issue. Rather than silently removed, it's more like some bug was introduced. macOS 26 is still Unix 03 certified.

I see macOS 26 listed under the Unix 03 standard. Was macOS 27 listed and then removed? The submission title + site contents combined do not give enough info to assess what exactly was silently removed.

> I also think this "paper then code" approach is now obsolete. The modern way, in AI-assisted workflows

To call the approach obsolete and refer to a "modern way" to use an LLM seems like a stretch when critiquing the approach used by a frontier lab a month ago.

Again, really sorry to derail the conversation. What you've created looks amazing! I do typical use the command line for mail (mblaze), but I'd certainly consider giving your client a spin if I wasn't on macOS.

The ad was for a 32TB portable SSD for $200...

https://wccftech.com/customer-buys-32tb-external-hard-drives...

Went to watch the Youtube intro [1] for Penguin Mail and was greeted by the most egregious scam video. Didn't click the ad, but simply went to the domain (usbest1.com) and it's just chock full of AI generated content hoping to scam you out of your money. Going through the checkout flow, it even asks you if you want to tip your driver (presumably for delivery).

Sorry for the off-topic, but Google (and all ad networks) should be legally required to vet their advertisers. That's gotta be the worst I've seen.

[1]: https://www.youtube.com/watch?v=0PyJCsw1FSE

See this recent paper: Quantization Degradation in Large Language Models: A Signal–Noise Perspective [1].

  We observe that such degradation varies substantially across these factors: 4-bit quantization usually preserves performance, 2-bit often causes broad degradation

This repo uses 2-bit quantization and removes some of the experts for its smallest fastest model. Make of that what you will.

[1]: https://arxiv.org/abs/2608.08188

Not sure why you are being downvoted, but this is definitely a limitation of current models and difficult to optimize for. Anything you are not specifically optimizing for is essentially unconstrained: the model could learn it by pure chance, but the likelihood is exceedingly low.

Techniques, like the newly announced RL-XAR from Meta [1] are being developed that will likely improve reward models and guide RL training to optimize for metrics like software architecture that are hard to verify otherwise.

[1]: https://facebookresearch.github.io/RAM/blogs/unslop/

I truly think it's safetyism. People seem to have a lower and lower tolerance for risk as society gets safer.

Since governments are run by people, this safetyism has become enshrined in law, typified by nonsense like criminalizing free-range parenting and mass surveillance.

It's pretty telling that even with RL post-training the big labs have essentially made little progress on the hallucination rate of models. The issue is fundamental to the current paradigm, contrary to humans.

GPT-6 Astra (max) has a hallucination rate of 51% and Claude Opus 5.5 (max) has a rate of 59% according to Artificial Analysis [1].

  AA-Omniscience Hallucination Rate (lower is better) measures how often the model answers incorrectly when it should have refused or admitted to not knowing the answer. It is defined as the proportion of incorrect answers out of all non-correct responses, i.e. incorrect / (incorrect + partial answers + not attempted)

Full speed ahead like an idiot savant trying a thousand different possibilities, though half of which are without basis in reality.

[1]:https://artificialanalysis.ai/evaluations/omniscience#omnisc...

> So if you happen to be able to make ␛]8;;https://example.com/"/...␇ appear in the job logs —4 which you can, either without even having an account, by sending a patch to a public mailing list with continuous integration turned on

The fact that this could be triggered so easily — just sending a malicious patch to a mailing list — is pretty insane.

Here are a few:

* Forming a long lasting memory from a single encounter

* Speed of adaptation to new environments

* Deft manual dexterity, like playing the violin and needlework

* Competing in a triathlon (requires running, biking, and swimming)

* Having a single agent that can do ALL of those things and more

Humans understand and perform tasks in many more than just the few modalities current model providers have focused on. I'm not discounting the utility of current models in the limited domains they operate in, but they are no where close to being generalists like a human or other animals.

They are simply looking to steer AI policy discussions. It's basically an in-house think tank. It has all the trappings, especially with ominous prognostications like:

  Today’s AI systems have impressive capabilities and the rapid pace of innovation suggests we’re now approaching artificial general intelligence (AGI), a system that exhibits all the cognitive capabilities of the human brain.

Not sure we are anywhere close to all the cognitive capabilities of the human brain... not even a severely lobotomized one.

I'm sure they would: https://tbreak.com/omarchy-m-apple-silicon-linux/

As another student who did the same, I similarly regret it, but also realize if I were put in the exact same scenario I would have behaved identically.

I literally had to retake trigonometry, while simultaneously being advanced into pre-calc, because my teacher saw that I aced the exams, but said I had to do the homework.

I ended up doing well enough: I dual-majored in CS and Math in undergrad and then completed a PhD in CS later in life, but I do sometimes think I could have achieved even more had I known better.