bhelkey

A "coherent" is not the same as "comparable to non-self-published authors".

Early LLMs were trivially non-coherent. The stories it wrote should shift constantly. The story could start on the moon, then the character could drive to Madison, Wisconsin.

A more obvious way to see this is in video generators. If the "camera" turns 180° twice, we often look at a completely different scene.

Can current models write an entire coherent book? I don't know. I have never wanted to generate such a book. But "do I like this book" is not a good test for coherence.

There is a quote from a CGP Grey video from a bit over a decade ago, "There isn't a rule of economics that says better technology makes more, better jobs for horses. It sounds shockingly dumb to even say that out loud, but swap horses for humans and suddenly people think it sounds about right." [1]

A couple hundred years ago, something like 70% of the population worked in agriculture. Technology replaced almost all of these jobs.

So far, we have reimagined old professions and created brand new ones. Is this ability of ours limitless?

[1] https://m.youtube.com/watch?v=CMFj75kBQlU

>Americans pay for it with tax dollars and personal data.

From the privacy policy[1], "Our AI providers operate under Zero Data Retention (ZDR) agreements and retain none of your prompts or the responses they generate. America.gov’s own response cache is separate and is described in Section 5.

Our AI providers are contractually prohibited from selling your information, building an advertising profile, or training their AI models with your data."

[1] https://america.gov/privacy-policy

Do you see the terms, "10x", "20x", or "25x"? If so, where?

I don't see any of those terms. Not even on the signed in page to upgrade to to a different Pro plan.

The plans I see are labeled, "Pro 100", "Pro 200", and "Pro 500" with the number corresponding to the price in USD.

> See the pricing page for a comparison of the features and usage included in each plan.

Does anyone see the usage limits? I don't see anything indicating how much usage is in each tier.

I checked:

* https://chatgpt.com/pricing/

* https://help.openai.com/en/articles/9793128-about-chatgpt-pr...

* https://chatgpt.com/#pricing (signed in plan upgrade page)

They are suggesting that a Democratic President could label Palantir a supply chain risk.

If:

1. The designation of Anthropic as a supply chain risk is politically motivated

2. We collectively accept that is okay for the US President to bar all defense vendors and contractors from working with an American company for political reasons.

What defense related company(ies) would a Democratic President have political reasons to ban?

It has been reported that Palantir and its founder have a history of political contributions to the Republican party.

Instead of Republican presidents driving left leaning companies out of defense and Democratic presidents driving right leaning companies out of defense, it would be better if neither party used supply chain risk as a political weapon.

Did tariffs raise prices? Or did prices stay the same because "if the market demand is such that it allows them to raise [prices], they'd already have done it"?

There is no lock in. Anyone with sufficient budget can download the weights for GLM 5.3. On OpenRouter, I count 29 different providers for this model [1].

In terms of purely local LLMs, one can run GLM 5.3 flash on a beefy workstation.

[1] https://openrouter.ai/z-ai/glm-5.3

Agreed, move selection was not great. Notably, it should not have allowed nxe7+.

However, the pawn was defended by the queen and it took a forced queen trade to unlock the move.

I have seen much worse blunders from human players. And, I have made much worse blunders.

It looks like it played a fully legal game of chess with one exception, it said "rxd1+" (Rook takes D1 with check) instead of "rd1+" (Rook to D1 with check) on move 29.

I would say this did a really good job of playing chess. It moved the pieces consistently and traded pieces when required.

This is worlds away from the frontier ~1 year ago where models would hallucinate pieces into existence.

Apple's announcement:

>Live Rewind, which can help users recall what was just said in conversation. A double press of the Digital Crown shows the previous 15 seconds of a conversation as a text snippet...Users can ask Siri about the content of the text or save it to the Siri app to revisit later.

The two fold format is much more popular than the three fold format. Samsung actual discontinued the Samsung z trifold [1].

Making the first folding iPhone a two fold format is a very prudent decision.

[1] https://www.samsung.com/us/smartphones/galaxy-z-trifold/

Will the watch only record conversations in public? There is nothing I see in the press release about that.

I (again) am not a lawyer but I strongly suspect that one still has a reasonable expectation of privacy even if the person they are talking to is wearing an Apple Watch.

There are two different features:

"Siri Recap" records throughout the day without the user needing to constantly enable it. It provides AI summaries of the recorded conversations including direct quotes.

"Live Rewind" shows an exact transcript of conversation but only of the 15 seconds immediately before the user activated it.

There are two different features:

Siri Recap records throughout the day without the user needing to constantly enable it. It provides AI summaries of conversations recorded including direct quotes.

Live Rewind shows an exact transcript of conversation but only of the 15 seconds immediately before the user activated it.

>Outside of phone calls, you are generally allowed to record anything you can legally hear

I am not a lawyer but I do not believe that is true in any state except Nevada. Two-party consent almost always includes in person conversations. California Code, Penal Code 632:

>A person who, intentionally and without the consent of all parties to a confidential communication, uses an electronic amplifying or recording device to eavesdrop upon or record the confidential communication, whether the communication is carried on among the parties in the presence of one another or by means of a telegraph, telephone, or other device, except a radio, shall be punished by a fine not exceeding two thousand five hundred dollars ($2,500) per violation, or imprisonment in a county jail not exceeding one year

https://codes.findlaw.com/ca/penal-code/pen-sect-632/

This is a new feature. It is an always on recording of conversation.

>Fable 5.1 will cost an estimated 25% less than Fable 5 for typical workloads, wherever usage is billed by token.

Firefox declared that they intend to keep MV2 support[1].

[1] https://blog.mozilla.org/en/firefox/firefox-manifest-v3-adbl...

This calls the top coding model for the Apple M1 Pro: qwen2.5-coder-7b, a model released September 18, 2024 [1].

I question this choice. Coding models have improved significantly in the past 2 years.

[1] https://qwen.ai/blog?id=qwen2.5-coder

Are you familiar with Ollama [1]? It is a particularly easy to use tool to download and run local models. They sort models recent popularity and specify size for the various quantization levels.

I would try using ~1/2 your available ram and iterate from there.

If you have 32GB of RAM, I would give Qwen 3.8 a try. All you would have to do is run "ollama pull qwen3.8:27b" then "ollama run qwen3.8:27b". If you have 16GB of RAM, I would try Gemma 4.

[1] ollama.com

Why did you use 1-bit quantization vs 3-bit quantization?

It looks like the 3-bit requires 90 GB[1] which, I imagine, would fit within the DGX Spark's 128GB of unified memory.

[1] https://unsloth.ai/docs/models/qwen3.8-next

He often promises incredibly ambitious things that don't happen. But, with a fairly reasonable frequency, he promises incredibly ambitious that do happen.

For example, SpaceX is responsible for a massive spike in orbital launches [1].

[1] https://ourworldindata.org/grapher/yearly-number-of-objects-...

> But the rare books under discussion are closer to the end of their life and less likely to have been already digitized.

It is not at all clear that this is true.

The number of books published every year is growing rapidly. According to Bowker the number of books published in the US every year has increased ~15x in the past two decades [1].

Because of this, I suspect that the median age of the books we are discussing is below 30 years.

[1] https://www.writercosmos.com/blog/how-many-books-published-p...

> it would be easier for them to distribute the work after the copyright of the work expires

Copyright does not expire for a very long time. Harry Potter and the Sorcerer's Stone was released ~30 years ago in 1997. It remains protected for the duration of the life of the author (J.K. Rowling) plus 70 years.

Given actuarial tables from the UK[1], this works out to be around ~95 years from now (~2120).

[1] https://www.ons.gov.uk/peoplepopulationandcommunity/birthsde...

Potentially I am misunderstanding you.

My point is that we already pay $2 Trillion dollars for existing single payer healthcare (Medicare/Medicaid). I think it's fair to call that extremely expensive.

You say that universal single payer healthcare would have a huge price tag, my point is that we are already paying a huge price tag ($2 Trillion) but just not getting universal single payer healthcare as a result of this spending.

My point is that we could have universal single payer healthcare without raising taxes if it cost less than or equal to ~$5800 per person per year.

UK spends ~$4800 per person per year on their single payer healthcare.

Note: This just considers existing taxes spent on existing single payer healthcare. This doesn't consider other healthcare spending. The majority of the US get's their insurance through employment-based plans. Some pretty reasonable percentage buy their plans directly through the marketplace. And some pretty reasonable percentage don't have health insurance.

> universal health care reduces aggregate costs, but the public cost in aggregate is enormous. This means large new taxes for everyone.

If the US could spend healthcare dollars as efficiently as the UK, the US could pay for a single payer healthcare system covering every resident using only existing Medicare + Medicaid spending.

Is the US willing to reduce the price of healthcare? I observe a large appetite in congress to increase the visibility into US healthcare spend. I observe rather more limited effort spent to use this data to reduce the aggregate costs (e.g. negotiating drug prices [1]).

The math:

The US spends $2 Trillion dollars on Medicare + Medicaid a year [2][3]. $2 Trillion dollars / a population of 342.7 million [4] = ~$5800 per resident per year.

NHS (UK's universal healthcare system) costs ~£3,500 or ~$4800 per person per year [5].

[1] https://www.kff.org/medicare/key-facts-about-medicare-drug-p...

[2] https://www.cms.gov/data-research/statistics-trends-and-repo...

[3] https://www.kff.org/medicaid/medicaid-financing-the-basics/#...

[4] https://www.census.gov/popclock/

[5] https://www.bbc.com/news/articles/cwy7zvp5xrqo

There are not enough physicians because we have limited the number of physicians. This was done, in part, by limiting the number of residency positions.

[1] https://petrieflom.law.harvard.edu/2022/03/15/ama-scope-of-p...

Who said anything about outlawing private insurance?

The UK has a single payer healthcare system but folks may purchase (or obtain through work) private insurance.