sheepscreek

Hey Thariq. Don't dwell too much on the negativity here. It's a hard act to balance and I'm sure the frightening pace has only made your job harder.

I think there might be a case for multiple CC release streams, same as Chrome - a stable that is infrequently updated, a beta (which would have been the right place for this feature to land) and maybe even a canary.

This way folks can choose the right one based on their risk tolerance.

AFAIK there’s no ZDR with Claude models accessed directly via Anthropic. You’d have to go through either Google Vertex, Azure or AWS for true ZDR (at least legally/on paper).

I’ve had great luck building SwiftUI apps with GPT-5.5 (now GPT-5.6 Sol), Opus 4.8 and Fable 5. I’m just offering another data point, not suggesting on the effectiveness of DSLs for this.

One line of thinking can be that frontier models are already powerful enough to brute force through it, but this might be a stronger indicator for success for smaller models.

Agreed. I think we’re entering an era where some level of specialization for general LLMs is a good thing. Particularly between tuning for agentic use cases (where you want agency with a ton of guardrails and control) and writing which is more creative - you want the model to take the occasional risk and not sound like a monotonic robot. Having trained models first-hand, I can see the distinct use-cases clearly that are odds with one another.

I wonder if they are double-counting Anthropic's leased capacity from SpaceX under SpaceX again.

I don't think it even matters. Because noone will continue to use an LLM that doesn't work well for them, whether or not it has a good bench result. So for their own sake, the correct representation can actually win them some loyalty:

eg. Model X is weaker than Fable, but competes well with Opus/Sonnet and costs 1/5th as much etc - something similar playing out with Grok 4.5.

Moneyshot:

The three bacterial strains that successfully induced tumor regression (E. americana, C. portucalensis, and E. ludwigii) were all identified as facultative anaerobic bacteria.

This finding is consistent with established principles of bacterial cancer therapy, as anaerobic bacteria possess the unique capability to selectively accumulate and colonize within solid tumors due to the characteristically hypoxic and immunosuppressive tumor microenvironment.

This selective tumor colonization likely enabled efficient intratumoral bacterial proliferation and, in conjunction with activated immune cell responses, contributed significantly to the observed tumor regression phenomena.

Said in other words, the tumours created the ideal environment for anaerobic bacteria to multiply. Eventually causing a reaction from the body’s own immune response (which ignored the tumour but successfully detected the bacterial growth).

So one of the reasons this worked well was that the bacteria acted as a target for the immune cells, and they proliferated inside the tumour thus weakening it.

Yes - they create new/cheaper products for a different consumer but not at the cost of their margins. Vision Pro may have been the only device in recent history that likely had slim margins (if any).

It's easy to blame datacenters, but there are many factors at play here.

Henrico County currently has 37 data-centres with ~2 gigawatts capacity (expected to reach 3 gigawatts).

Apparently, 1 MW can power approximately 834 homes annually. So 2 gigawatts would be closer to powering > 1.6 million homes.

Surely, that kind of concentrated demand is going to affect electricity distribution costs for everyone, which is what we are seeing now.

I’m a complete layman to this field, but what the article did say was they’re hopeful that AI/ML can help develop a model that can pull out information such as the scattering caused by RBCs (which is present in the large volume of data gathered by the probe but is too weak to be used for manual techniques) and turn that into meaningful visuals. That’s gonna require a ton of data and that is exactly what they are trying to gather now with what they have built so far.

The message I’m getting is that Apple will never compromise on its healthy margins. If something becomes basically unaffordable for their target market, they’d cut the production and even discontinue the product, than take a hit on margins. Their business model is refreshingly simple.

Speculating here - “effectively” cooling the CPU and GPU materially using this technique at datacenter scale may have never been done. Those things than run hot, easily crossing 100C. So the loop is doing a lot of work to keep them stable at 55C.

The innovation may be in the speed or volume flow of the coolant through different parts of the data centre to regulate the temperature. And of course, redesigning every component to be compatible with this fan-less design.

I think it’s only possibly because NVIDIA is much more vertically integrated than ever before.

I think it’s more than that. Piecing together the perspective of a few commentators in this post - it’s plausible Anthropic is trying to shift the narrative from US vs. Rest of the world to US vs. China.

In other words, they want to sell Fable or future more powerful models to rest of the world (presumably all future models are going to be more powerful than current gen). One way they can sell this is to the government is by scapegoating China (which is their primary concern anyway).

This is working on the presumption that non-US companies form a material portion of their current revenue.

So I think the takeaway here is, this is a super fast companion model to larger models, that reasons quickly. Perhaps this technique can be used to train a highly optimized reasoning "expert" in MoEs.

The initial motivation for this was likely to thwart any competition. Already Anthropic has accused some companies of organized distillation efforts at a massive scale.

Back when I used antigravity, it used to show the reasoning intact - at least for Gemini Pro 3.1, and likely for Claude Opus 4.6 (not 100% certain about it). I have some recollection of stopping the models mid-turn when they started going astray.

As a power user, I find reasoning fascinating to read and genuinely useful at times. Probably not that useful for 80% of their base.

You don't need/use pitbulls, but what if you (and many many others) wanted and needed Fable?

I for one was late to the bandwagon, and when I had the use-case for it - the govt pulled the rug. So yeah, I'm a bit salty about the whole endeavour.

I will also say that the security concerns are probably very real (and they have been from the day ChatGPT-3.5 came our). I guess I can be salty about it and still be wrong from their perspective. The govt likely understands the fragility of their infrastructure better than us and is likely aware what this could unleash for their systems.

As an extension to this, in this world - most typical things humans need for survival will be made extremely cheap and abundant. Universal income would give every individual the means to afford them. Every person on earth (or at least in certain societies) will be able to live comfortably (not luxuriously, but comfortable - ie. survival becomes commoditized). People would only work to have a purpose.

I have been discussing a similar thought experiment in private circles for about 3 years now. It goes like this:

If the world population hits a ceiling (or starts to shrink), and corporations get more massive - reaching every citizen on earth, how can they continue to grow?

Assuming the typical pathways of acquisition/merger have been exhausted. Now you're looking at a world of abundance, and corporations must invent ways to generate demand.

So what if, robots could be state owned and be treated as a special class of robot citizens? They could earn a wage (or some kind of credits) and spend them for benefits like upgrades or repairs. Not just humanoids, all robots. With this, one could essentially create infinite demand-supply channels.

I believe this was the point they’re making.

Turns out, the friction I felt around picking one thing may have actually been beneficial. Perhaps it was actually helping me stay focused. Even if it cost just a bit of extra time before I sat down and worked.

They regretted it.

I have brown eyes, still hurts to drive at night. Canadian situation is just as bad as our neighbours down south. Now I long for the days of being blinded by yellow headlamps again.

This is actually a good thing. Google's official mission is "to organize the world's information and make it universally accessible and useful."

Now if that information is BS and cannot be relied upon, that’s really bad. Leading people on and not delivering? Honestly, Google themselves should have been on it and not the German government. It’s a bad look for them.

Aside: I’ve noticed their AI mode is pretty pathetic for troubleshooting something. 50% of the times the first response is riddled with inaccuracies and mistakes. Repeat prompting is absolutely necessary (so do not expect to one shot anything).

Also I will admit that I still find myself using it because I’m lazy, and it’s easier to talk to AI to get the right answer. Searching organically is hard these days with the volume of content having gone up exponentially.

I love this term - I think it beautifully describes the direction that at least, YouTube is heading towards. Take for example, this racket where a channel copies popular (non-kids) creators’ parody work, splits the screen in half with the content on left, adds a completely random DIY type video on the right half, and lo and behold its content for kids who are too young to know any better[1].

Another one: AI voiceovers on videos taken from Asian apps, with some made up emotional story, followed by “if you love your mom, like and subscribe” - which kids (< 8yrs) actually do![2]

Or for that matter that YouTube makes it so hard to block channels and impossible to unblock specific channels (at least for kids). The platform has been unwilling to do anything about it for years. I suppose maybe this isn’t the best example but it’s definitely along the lines of a corporation prioritizing profits over all else, especially disregarding the wellbeing of their users.

1. https://youtu.be/VF4V7bRjjdo https://youtu.be/UoGuLabqgrk

2. https://youtube.com/shorts/B2ZNFiix8JA https://youtube.com/shorts/0eYYKRRcYrA

It’s exceptionally fast at it too. I love using it for looking up things where recency matters.

0.1% is within margin of error. Depending on the performance boost, it might be worthwhile taking a minuscule quality hit.

I think it’s meant as “for the first time in history..”. Not today in particular, but as a milestone.

I was commenting specifically on this bit:

Once the chip shortage from AI datacenters bubble pops, we could see even better SoCs from Intel, AMD, and even Qualcomm and Nvidia could join the ARM laptop battle in a serious way.

EPYC (the 2nm AMD chip being produced at TSMC Arizona) is still going to support the datacenter demand, not consumer devices. You and I are still screwed. Apple is the only behemoth, IMO, that wields significant power against other trillion dollar companies. Not Acer, not HP, not even Dell (I don’t think). This is my personal opinion, I don’t have specific facts at hand to back it. Just a strong intuition.

But hey, govts might step in at some point and say - we need to put a cap on how much supply data-centres can buy. Since computers and phones are the backbone of modern society. There may be some rationing happening down the line.

FYI - DeepSeek has NOT announced its own coding platform. That app is an independent project. It says so in the footer as well:

“Independent open-source project · not affiliated with DeepSeek”

The (memory) chip shortage saga is not going away for a few years. Most fabs are going to be capacity starved. Apple will happily pony up billions to TSMC to set up a new plant in exchange for exclusive capacity. No other laptop manufacturer can do this. This will put them in an even more advantageous position. In all honesty, the Neo couldn’t have arrived at a better time for them.

They aren’t making new paid products for a growing user base. They are continuing to support their paid Linux user base.

I never said, or even implied that they are creating a new product. They are dropping the “free tier” or licence, if you will, because either a) they are starting to see this tier being used for commercial applications b) they are maintaining a completely separate code base/product for free users and don’t want to support its development anymore.

If it was b) they could have chosen to open-source it.

Any other reason that I can think of is probable but unlikely - like, there’s a non-zero chance they could be trying to beef up the revenue on paper, before selling off the FPGA unit, to focus on servers and GPUs.

“Until now, it has been available for free on both Windows and Linux”

If it’s any consolation, it wasn’t and still isn’t available on macOS. Also the part about Linux having a “small user base” made me chuckle.

That’s the opposite of what I’m observing. If they wanted to save costs, they would have dropped Linux support altogether. But instead, they are making it a paid benefit. It can only mean that their Linux user base is growing, ie. more commercial operators are turning to Linux. Still, there are much better ways to handle this without alienating your user base.