ttul

That's not the point. Often in medicine, the downside of using one thing is that you're not using another thing that would be safer and more effective. When people chase results from unproven therapies, often it means they're avoiding doing something else that is known to produce a result - even if the result of the alternative is not as spectacular as the unproven therapy promises to be.

But speaking of creatine itself, while not inherently toxic, consuming large doses can cause serious indirect harms, such as triggering falsely elevated kidney markers on blood tests that lead doctors to misdiagnose organ failure, unnecessarily cancel crucial medical treatments, or order invasive biopsies. Furthermore, overloading the body with creatine while taking kidney-stressing medications (like heavy doses of ibuprofen) or having an undetected renal condition can cause actual acute kidney strain, while the massive shift of water into muscle cells can trigger severe dehydration, heat exhaustion, and electrolyte imbalances during intense heat or exercise if fluid intake falls behind.

But, I don't want to belabour the point about creatine. Rather, I want to focus on the negatives stemming from NOT doing other things that ARE proven to be safe and effective. People tend to think creatine will solve a lot of their problems (for instance, they may believe it will fix their feeling of being "low energy). But by avoiding things that might actually help their problem get better, they're persisting a poor state of health.

What value is there to an informal conclusion when the topic is as important as intelligence? The downsides to creatine if taken without good scientific backing should not be disregarded, yet that’s exactly what many ordinary people are now doing because of “informal conclusions” like this.

If the weights are open, censorship can be easily trained out of the Chinese models. But if you’re sending your tokens to China, all bets are off!

Oh yes, I know it’s a harness feature.

Fable seems to be a larger model. It costs more to run and does not seem superior for _typical_ software engineering work. But for work requiring raw intelligence, perhaps its size is an advantage.

On the DeepSWE 1.1 benchmark (IMHO currently the most relevant and least gamed SWE benchmark), the cost-benefit is clear: 5.6-Sol on xhigh achieves a slightly higher score than Fable 5, but consuming half the tokens and at about 1/3rd the cost.

But, on the Artificial Analysis intelligence index, Fable 5 appears to slightly beat 5.6-Sol, albeit at 3x the cost.

When I am coding, I send tasks to each model to get multiple opinions and it can be hard to predict which model will “win” because the results can be subjective. OP’s task is at least quantifiable, which is great. But many SWE tasks cannot be quantified so easily.

I think ultra mode needs to be more clearly documented (or perhaps cautioned against!). Most devs - myself included - who saw “ultra” mode figured that it’s just a magic bullet that makes the model work harder and achieve better results. But, for many tasks, ultra mode is possibly worse and certainly more expensive.

It's kind of amazing that this took so long. On the other hand, this is just chapter 3025223 of the failed war on drugs and we can be confident that people will find something worse as an alternative.

A lot of the work that coding agents do requires tons of random access to files, which MacOS is particularly slow at. Not because the hardware is bad, mind you. Mac SSDs are amazing. It’s the file system. Secondly, MacOS supervises processes for security reasons and coding harnesses spawn tons of processes. There is a lot of CPU overhead embedded in that supervision.

And, make coding harnesses run 20x faster at many multi-processing and file operations by running the coding harness itself inside of Linux instead of MacOS…

You are not wrong, but the practical reality of local hardware is to be batch-constrained in comparison with a multi-user inference cloud. You will never be able to compete cost effectively in your home lab with a cloud that has >100M end users streaming millions of inference requests per second across a gigantic fleet of machines.

Can I run a few inferences in parallel on my Mac Mini? Yes. But put 1,000 Mac Minis in a datacenter serving 1,000 copies of myself? That's going to be more efficient.

Here’s the two main reasons why local inference won’t compete any time soon with the cloud:

1. Most useful LLM work is done in parallel. A Mac Mini can run one LLM inference thread at a time. The cloud can spool up dozens and spread that inference across efficiently batched operations over a fleet of hardware.

2. Faster inference hardware such as the chips from Cerebras and Groq cannot be run locally. But the advantages of running >5x the token throughput per thread can’t be overstated. Add in the multi-threading advantage and it’s a knock-out punch for local LLMs.

Local inference has a role: if you’re working with extremely private matters or you want an uncapped model that will talk dirty or generate NSFW photos, local is the only option. I think Apple and others will continue to also run a lot of useful workloads locally such as text editing suggestions, speech to text, text to speech, and image manipulation. As local hardware improves, these capabilities will get better too.

But, for most LLM work, the cloud will continue to dominate for a long time to come, if not forever.

I disagree with this assessment. Theo is nerdy and, yes, he has a healthy ego. But, he provides insightful commentary on his channel and he works very hard to present what he believes is the truth. Compared to many YouTubers, whose content is vacuous, Theo is mostly the real deal.

I’m guessing the OrbStack team will probably support MacOS native containers soon enough, with all their management goodness on top.

How much time have you spent with Fable? We're in new territory here. It does not create messes.

It’s not about Iridium. It’s about Iridium’s customers and partnerships. RocketLab hopes to launch their own satellites presumably and then can sell significantly improved services to them, without having to build a customer base from scratch.

And anyone who has done an introductory course in VLSI design would know that capacitance (coupling) is something you usually want to get rid of. However, all kinds of amazing analog circuits have been developed over the decades that exploit coupling effects. So, their idea is not outlandish at all.

What they are trying to achieve is to demonstrate that the coupling approach works in a simulated physics environment (O(n^2) as you point out) so that they can then build CMOS circuits that create actual oscillators and then let the laws of physics do the computation. This is a very bold vision!

It's pretty great that you are providing the undistilled model on day 0. Here's a pro-tip: With Flux.2 Klein, someone created a turbo slider LoRA - basically a diff of the turbo 9B model vs. the undistilled 9B model. What's great about this LoRA is that you can sample using a heavier weighting of the undistilled weights during early sampling steps and then finish the sampling off with mostly the distilled weights. The result is a better "finish" (taking advantage of the distilled model's refinement for image quality) without sacrificing the undistilled model's greater ability to adhere to the prompt, because the undistilled model doesn't have to devote its weights so much to looking good.

This is a massive technical report for an open weights image gen model. As someone who has followed this space closely, it’s really cool to read about the behind-the-scenes experimentation and effort that went into the final product. I hope you will release some of the find tuning tools so the community can experiment with them as well and really push what the model’s capable of.

Good points. I would refine my comment: Canada has isolated examples of excellent megaproject execution, but lacks a system that reproduces that performance reliably. The nuclear announcement is credible only insofar as Canada can replicate the OPG/Bruce fleet-program model—stable authority and funding, standardized designs, repeat teams and accumulated learning—rather than its usual election-cycle, one-off megaproject model.

I dearly hope that we're not going to see that model again, but maybe that's likely.

You're right: you didn’t claim a lack of engineering competence. I introduced that distinction. Sorry. If the question is whether the Canadian society can build reliably, then governance and political instability count as part of its overall competence, not as external excuses.

I would agree with you that political volatility exists and stifles an otherwise decent country. It's very frustrating.

Canada is not an infrastructure “joke.” It is a country with some world-class delivery organizations operating inside a political system that too often destroys continuity. Relative to the G7, that makes it mediocre and volatile, not uniquely incompetent. And, in nuclear specifically, probably no worse positioned than its peers, though the ten-reactor rhetoric is substantially more ambitious than the underlying commitments at this time... (not surprising - it's a politician making an announcement, which is something of a prerequisite for making a "real plan" anyways).

As a Canadian, I think Canada’s primary hurdle is not a lack of engineering competence, but rather political volatility. Projects like Calgary’s Green Line often suffer from shifting scopes, fragmented authority, and delayed funding. Conversely, the recent Darlington nuclear plant refurbishment finished early and under budget. This proves that Canada can successfully execute megaprojects when planning is front-loaded and standardized.

Another comment I'd make is that the Carney government is only just a bit more than one year old. They're writing a whole lot of new policy. Will they succeed more than past governments? Who knows. But, at least they're spending the majority of their political capital trying to build stuff.

My first year English teacher would be impressed. The lack of em dash characters in a paper would get you a failing grade back in 1995.

True. Z.AI ran that bench themselves and report 46.2, which is lower than GPT-5.5 and Opus 4.8, but crushing the other open weights models.

https://z.ai/blog/glm-5.2

DeepSWE “feels” like the right benchmark in comparison to Artificial Analysis indices and other coding benchmarks. And by their metrics, GPT-5.5 is still king in token efficiency, speed, and overall intelligence per dollar.

https://deepswe.datacurve.ai/

Fable 5 is cool and all, but we have not yet seen GPT-5.6.

Congratulations to the Intercom founders. This is a great result for them and the early backers of the company, for a product that I feel was executed brilliantly. We can criticize the usual enterprise-sales-y trajectory and enshittification that ensued after the brilliant opening moves, but at the end of the day, I think Fin reached a solid ending (no pun intended).

Is there any evidence that this actually takes place in a consistent, widespread manner?

I'm not sure whether you're aware of how racist this comment may sound to many readers here on HN, but you're definitely echoing the sentiment of many Canadians who sense that "foreigners" are taking away their jobs - a concept that has been leveraged by conservative politicians to stoke distaste for the current government. There is indeed a lack of youth employment in Canada, and the TFWP may be somewhat to blame for that. But labelling it as a problem with regard to a single group of immigrants strikes me as unhelpful.

There is actually a sensible way to do recruit foreign workers to fill jobs that locals for some reason can't fill, and it's just a few miles up north...

Canada's Temporary Foreign Worker Program (TFWP) is not limited by annual caps or lotteries. You just apply (as a company) by filling in a few simple forms and posting a job in the "Canada Job Bank" for a period of time to demonstrate that you genuinely searched for locals to fill the role and couldn't find anyone suitable. I've hired many people through this program to fill a variety of roles over the years, and all of them eventually became citizens too. Once you're on Canadian soil as a TFW, moving toward permanent residency is not very difficult if you're a skilled worker with enough "points" (based on education, etc.).

Some argue (perhaps correctly) that the TFWP suppresses Canadian wages and productivity growth by flooding the labour market with cheap staff from poor countries. And there is likely some truth to that. But when I hear how many hoops my US colleagues have to jump through with lawyers and such to bring skilled employees in, it boggles my mind. If the Americans were to implement a more modern temporary foreign worker program similar to what Canada has, you'd have to imagine the US economy would boom like it never has.

“Quick, guys! We finally made a profit and our ARR is up 80x in three months. Let’s go public before any of this comes back to earth!”