A_D_E_P_T

> "By Kpler's count, the GCC states are now moving about 24 million barrels per day out of the region, including volumes transported by overland pipeline."

Complete BS article. I guess Kpler is the propaganda mouthpiece.

The article double-counts "Hormuz Shuttle" ship-to-ship (STS) transfers and misreads pipeline capacity.

An example: If 14 mb/d is moving through the Strait of Hormuz, getting to 24 mb/d requires 10 mb/d to move via overland bypass pipelines. In reality, the only two bypass pipelines out of the Gulf have a combined physical ceiling of ~6.5 to 8.5 mb/d. (And that's when they're fully operational, which may not be the case today!)

Many more obvious mistakes in this "analysis." It's totally bunk.

> I have a growing suspicion that this is the Great Filter.

Ackshually... It can be our filter, but it cannot be the Great Filter.

The Great Filter has to apply universally, to every plausible society. If even a few are cooperative hiveminds, or empires ruled by an iron fist, or... whatever else you can imagine that doesn't squabble internally... then the Filter does not apply to them, which means that their absence is still weird.

The Filter has three plausible answers, IMO: (1) Some variant of the planetarium or zoo hypothesis; (2) every advanced society, universally, decides to transcend into data centers powered by rotating black holes and stable for trillions of years, or something similarly weird; potentially also New Physics, like waste heat to mass conversion; (3) the filter is in our past and abiogenesis is a lot rarer than assumed.

Cold emails are spam, unless those people specifically opted in. And it'll harm your product, probably... Blasting 50k scraped or purchased email addresses from a primary startup domain is one of the fastest ways to permanently brick any company's communications infrastructure.

Modern email providers track sender reputation primarily at the root domain level (your .com). Doesn't matter if you send via a service with a different IP than yours.

And the spam threshold today is 0.3% (that is just 3 people out of every 1,000 clicking "Report Spam"). If you're over the threshold, you're auto-flagged.

In short, you'd surely get severely penalized, and your host/registrar are also gonna get mad. Don't do it.

Don't spam people. Just don't do it.

> And we haven’t even got to the part where AI consumer pricing corrects to realistic models. Once that hits, companies will drop AI like a hot potato.

That ship has sailed. If you're a mid-sized company, with the resources of a mid-sized company, you'll always be able to download Deepseek's new Flash model and run it on your own hardware.

With all of the competition out there, prices arguably can't rise much further than they already have.

I am serious. I believe that the LLM is the defining technology of our time -- at least as important and as impactful as the internet itself -- and that Meta's opening of LLaMa has profoundly changed the landscape for the better.

I never said they were open source. Everything else you wrote is factually incorrect.

When you run a proprietary program (like Adobe Photoshop or a closed-source game) on your PC, you get a compiled binary of machine code. The high-level logic is stripped out, variables are obfuscated, and the EULA legally forbids you from decompiling, reverse-engineering, or modifying the binary.

With an open model, you get the actual Python/C++/CUDA code defining the exact mathematical structure of the model. And you get a weights file, which is an unencrypted, open array of floating-point numbers representing every synaptic strength in the network.

So you can inspect the exact mathematical state of the model at any arbitrary moment during inference -- in other words, you can study open models. You can fine-tune or quantize them -- in other words, you can improve them quite significantly! And much much more.

It opened the floodgates, though. It was a real paradigm shift. After LLaMa, everybody saw that you could release large open models without any negative repercussions, and DeepSeek et al. hopped aboard the wagon. LLaMa was the first popular model that was truly open to the global public.

Meta earned a ton of goodwill by doing this -- it was very brave, and very influential. In my eyes, it outweighs everything negative they've ever done.

The Google models are great "general purpose" and research LLMs, but they're relatively poor at coding.

The way I see it, you have two options.

1) Get on the Claude Max 20x plan and use Opus 5.5 for review/orchestration and Haiku 5.5 for grunt work. Haiku 5.5 is decent, cheap, and fast.

2) Deepseek 4.1 Flash in OpenCode.

You should also test MiMo v2.6. It's a little bit uneven, but might work well for whatever you're trying to do, and it's also cheap.

I'm the same way re: bottoming out my 20x Max plan in ~2 days. I'm having better luck lately, though. What helped was using Opus 5.5 for planning and orchestration, and Haiku 5.5 for all of the simple stuff. Claude doesn't default to this -- the default is that it uses the same model for everything -- you have to give it specific instructions.

White dwarf stars can be dyed just like (colorless) gemstones and (white) ceramics, then. It's an essentially similar mechanism. I wouldn't have thought it possible, but I suppose it makes sense.

Interesting!

For me it's large files in Blender (in some cases >400MB), mostly, where Astra excels. I gave it a look, but I don't know if the benchmark encompasses that sort of thing.

.step files are simpler pieces of straightforward code, so it makes sense that Opus and Sonnet 5.5 do very well...

That hasn't been the way to bet. DeepSeek and the other Chinese firms show no signs of slowing down, so if OAI and Anthropic halt releases and "pace the frontier" they'll just fall behind. (And there's a chance that high-end AI will become fully commoditized, which suits DeepSeek just fine.)

Besides, Fable 5.5 is reportedly releasing within the next few days, probably on Tuesday...

OpenAI's models -- and even Google's Argon -- have particular niches where they shine. For GPT-6/6.1, it's 3D modeling; Astra is hands-down the best at making .step files, Blender files, etc. For Argon, it's natural language writing and everything that hooks up to Google, e.g. search and maps-based itinerary planning.

Anthropic and DeepSeek (+ GLM) are definitely solving coding, though!

They're all quite different tools.

I can't take any credit for it. I believe Terence Tao coined it first. He also wrote a bit about it here: https://mathstodon.xyz/@tao/117237320796901560

I don't know about cryptography, but if you try anything in organic chemistry or biology, you virtually always get flagged and downgraded.

As this is going on, Anthropic has opened up their own biology lab.

> https://www.reuters.com/world/anthropic-quietly-sets-up-biol...

There is very much a situation where people in frontier labs have access to exceptional capabilities that are being kept from the rest of us, and they're using it to strip-mine technical and scientific problems.

The metaphor was, in my opinion, far too badly strained to work.

Such a weird take.

There's a library of babel simulator here: https://libraryofbabel.info/

Just hit "random." It's always gibberish. The odds of getting something that isn't gibberish are astronomically low, like winning the lottery.

The actual library Borges describes has 25^1312000 books, the very vast majority of which are utter nonsense. (As the simulator makes clear.)

Yet with AI, we're getting an abundance of very credible -- and, as Tao himself notes, frequently correct -- mathematical proofs. In these new AI math results, the ratio of signal to noise is high, which makes our current situation the exact opposite of the library, which is pure noise where signal emerges only by random necessity.

There was an interesting passage in the article:

> Today, the mathematicians are weeping, because the machine smashed their tower first. But tomorrow, it may be the physicists, the economists, the illustrators, or the writers who weep over their fallen blocks.

Isn't it extraordinary that these models still can't write well? I mean, they write passably well, but they all have a highly characteristic style that you can spot from a mile away. These language models can crack the toughest open problems in math, but they haven't yet cracked human-like writing, especially long-form writing, without exposing themselves. This is very very odd, IMO.

> If you try to visualize this in action, replace economic migrants from Latin America with spies from China. And you quickly see there’s nothing inherently wrong with the idea.

Is this a joke?

This probably has more to do with US government pressure; they clearly don't want refineries or energy infrastructure hit in Russia.

gemini.google.com with a $20/month Pro membership.

I'm not the poster you responded to, but I can tell you that historically the Gemini models have lagged behind Anthropic (the clear leader) and OpenAI's best models where coding in particular is concerned. Opus 5.5 is, right now, the consensus best coding model.

That said, I'm extremely impressed with Argon thus far. I take back everything bad I've ever said about their AI strategy. Argon is very good and very fast. It might be competitive at coding, but it's too early to say.

Totally agree. This is my new favorite general-purpose/non-coding model. It's surprising how good this thing is...

You're right on this point. But there are ways to build interfaces that sidestep this.

Run a sanitized VDI with no local storage --> offshore independent entity (e.g. a Hong Kong Company) that holds the AI service contract and API license --> GLM-Legal.

Somebody ought to set this up for pro se work.

You got it wrong. If courts and attorneys in the US can't run "discovery" on Z.ai servers or compel them to do anything, and if Chinese authorities don't cooperate with the US legal system as a matter of principle, then what you functionally have is super privilege.

> These models lack the taste that makes for effective product decision making. It is cheaper than ever for a product manager to say "Someone complains about this on reddit, let's implement it", while the tasteful leader says 'No, that's not what our product is. Making that means we commit to it for a long time and our customers really want {something else we already do, but not perfectly} instead.'

This is a very good point. It's now easy to just do things, so discernment has become more valuable than ever. If you leave things to Opus 5.5, it'll just say that you need to implement every product-bloating change, and "here let me take care of that for you".

This is going to sound crazy, but, at the time, they were spooked by Neal Stephenson's June 2019 book The Fall -- where the internet was broken by the open-source release of something akin to a language model that generated endless quantities of spam. They thought that GPT-2 might be that invention, so it worried them.

Then, a few years later, they learned that Google and Anthropic were planning on releasing LLM chatbots to the public, so they hastily launched GPT-3.5-based ChatGPT as a "dev preview" to get ahead of them. Their original plan was to finish training GPT-4 before launching anything.

Commercial considerations -- and public glory -- triumphed in the end.

Okay, but as I just put it in another thread (which was in "new" and never made it anywhere near the front page):

All we have to do is assume that things progress in a linear way.

Do you remember GPT-2? It was a bizarre curiosity, but was nevertheless considered a huge advance in late 2019. Practically useless, but interesting.

It grew, and three and a half years later we got GPT-4, which was exceptionally capable on release.

It grew still further, and three and a half years hence we now have GPT-6. This is already more intelligent, more imaginative, more disciplined, and far more erudite than the vast majority of individual humans. If you have an intellectual skill -- if you are, say, a mathematician or a chemist -- GPT-6 also has that skill and is functionally your peer. And then it has many other skills that you lack; if nothing else, it is superhuman in breadth.

Assume a steady rate of progress. In three and a half years, we're going to have something that is genuinely a general superintelligence, and we're probably going to see recursive self-improvement begin to kick in (though subject to physical constraints on compute, etc.)

Now is the time for an RSI start-up, I guess!

Anyway, all you have to do is assume that things don't plateau. They have not plateaued to date; if anything, progress appears to be accelerating, and people are starting to panic that there are no brakes on this thing.