riknos314

This strategy only works as long as China continues publishing weights. Domestic training capabilities are one way of mitigating that risk.

Good! The goal isn't preventing chatting, it's providing separation from "the algorithm" (a dopamine casino, where you lose attention instead of money)

> that allows us to compare their performance. Opus is considerably better.

In coding? Software architecture? Math? General knowledge?

The diversity of model use-cases is so broad that comments about "model x is better" without any context are largely useless

Initial impression is that the currency selection didn't work (changing between USD/EUR on the selector didn't change the currency being displayed in PLN)and I quickly lost interest due to lack of monetary context

LLM seems to have become synonymous with Generative Transformer architecture.

While this model may share much with GPT-style models on the encoder side, it clearly has a different decoder architecture. So is a high-parameter count language model an LLM even when it doesn't have a GPT-style decoder? The definitions are in flux.

Has LLM become so synonymous with Generative Transformer that other high-parameter count models that interpret language need a different name?

For all we know this might be a non-language-generative transformer e.g. a transformer where the decoder produces confidence scores rather than language. Please provide more likely architectures if you know them, I'm genuinely curious.

This is likely still an LLM (in the purest definition of a language model with relatively many parameters) since the inputs are natural language, just not a generative LLM as the output is something other than more language.

Public communications from government entities really should be legally required to be disseminated on public forums (ideally properly hosted on .gov domains for easy verification, with RSS feeds) Cross-posting on other platforms should be allowed (with backlink to the .gov), but the .gov should be authoritative.

At least in the US most places that sell gift cards require ID when purchased in cash

Halved would be a 50% percent reduction.

0.15/0.22 ≈ 0.68, meaning a roughly 32% reduction on inputs. The 50% reduction is only outputs and cached inputs.

The author misses the difference between cashflow and profit entirely.

Selling a car today gives Tesla the full profits of that hardware production today. Running that car as a robotaxi means that Tesla eats the costs of the hardware today in exchange for a larger total profit collected over several years.

So operating a fleet gives the company access to more long-term profits at the cost of decreasing the bank balance today (negative cash flow), where selling the cars lets the company fill the bank account right now (positive cash flow) at the cost of limiting long-term profitability.

The decision to prioritize immediate cash flow vs long term profits depends on the financial position and overall strategy of the company.

also seeing 404

I'm not sure if the website allows for diffs against arbitrary tree states, but if it does than the diff space is completely unbounded, and the ram demand is theoretically infinite.

The amount of memory required to cache all possible diffs (defined as an ordered pair of commits) would likely be in the exobytes. At current ram prices that's easily a trillion dollars of ram to run that cache lol. Git focuses on making diff calculations efficient largely because the space of possible diffs is very expensive to enumerate.

The following from Claude: """ A diff between two randomly chosen commits usually spans years of history, so it's not a few KB — the tree itself is ~1.5 GB of text, and a multi-year span rewrites a large slice of it. Call it 100–200 MB per pair on average: 8.5×10¹¹ pairs × ~2×10⁸ bytes ≈ 10²⁰ bytes, or ~150 exabytes """

For anyone interested in the differences between dynamo (the early internal-only KV store described in this paper) and DynamoDB (The AWS service), Marc Brooker has an excellent writeup https://brooker.co.za/blog/2025/08/15/dynamo-dynamodb-dsql.h...

I'm curious if this stems from the focus on token efficiency. My thinking is that in order to use fewer tokens the model must converge to a likely path faster, meaning it must be more confident in making assumptions quickly and not second-guessing them.

Time to market also matters a ton. If they can start shipping chips <1 month after the weights drop that's much more compelling than if it's a 6+ month development pipeline.

Kimi K3 is a 2.8T model that's available at about 1/4-1/3 the cost of Fable from multiple providers on openrouter. The math doesn't seem wildly off.

Yeah but the lease is provided by Klarna, not Apple.

The proper server AI chips come on a board that isn't compatible with consumer pcie lanes and have no built-in fans. Completely different ball game in the ultra dense server world.

Buying the team and then actually making the models open is an interesting avenue for driving hardware demand (basically making openai models the new nemotron).

Unlikely but quite interesting.

You're welcome to break this down however you wish. I don't think there's any perfect measurement or stat here as every consumer values the various components of the combo differently. If you are a consumer that values the drink highly, then a dog-only analysis is incomplete. If you only care about the dog then you want to leave the drink out. Neither of these calculations is inherently more or less accurate than the other, they're just relevant to different preferences.

My original point is that by changing what metric and how much of the combo you look at the calculated loss can vary significantly: over an order of magnitude difference between the dog-only volume loss and the refilled drink calorie loss.

lies, damned lies, and statistics.

Soda isn't my preference and personally I'd probably skip out, but it feels disingenuous to completely ignore something included in the price when discussing the value of an item.

That's why I have percentages for both the dog alone and the combo.

I'd argue that the author's choice of volume as the metric to compare against is a poor one.

The costco hotdog contains 580 calories, approximately 110 of which is the bun. If the bun shrinks by 30%, the calories of the entire dog shrinks by 5.7%%. But also the $1.50 is for a combo that includes a drink of up to 270 calories, so the caloric content of the combo shrinks by 3.9% -- substantially less than the 37% claimed volume loss (which does not account for the drink).

EDIT - The drink also allows for a refill which would add another 270 calories bringing the total to 1120, and the caloric shrinkage down to 2.9%

When looking at protein the meal loses 1 gram out of 23, or 4% loss, which is less than inflation. Good for all the gym rats.

Edit 2 - fixed percentage calculations

> Whenever corruption occurred, we had to stop the control plane process on the shard while we repaired or restored the database. This was painful for tailnets on that shard, because their entire control plane disappeared during that recovery window.

Gotta love single points of failure...

As long as the linguistics are the issue rather than raw inability to understand the concepts being communicated, then the model is just being a bad communicator.

The code is the "how", the problems that people need solved are the "why". The code is merely an implementation detail.

For most problems solvable in code, there are literally infinite valid code representations of possible solutions (at least in higher-abstraction languages, the space is much more bounded at the assembly level). If you consider the subset of those representations that result in sufficiently performant execution, does it really matter which one is running? This is where most programmers will start mentioning clean abstractions and readability and maintainability, but those are concerns of making the code understandable for people reading and modifying that code. The people are still the driving force.

I'm not arguing for slop here - I definitely care about code quality - but it's important to stay grounded in the reason that it matters: people.

Aws is an infrastructure company that builds services on top of that infra to sell more of it at a higher margin.

Anthropic trains models on AWS's (and GCPs, and Microslop's) infrastructure, then skims margin off of selling inference also on the infrastructure owned by the other companies.

These are extremely different businesses.

The $200 sub is customer acquisition cost to hook devs that then become the marketing team trying to get their company to bring in Claude (at the highly profitable API price).

Effective jargon usage is understood by the target audience.

If the AI is communicating to me and can't select the appropriate jargon level, it's failing at communicating effectively.