wcoenen

What is "Retina RDP"?

In the page I linked above, Anthropic explains that they use the techniques from SynthID-Text, where a secret watermarking key is involved. Only Anthropic knows this key, so only Anthropic can produce and detect the watermark. (But they also promised to provide an API through which other parties can check the watermark.)

SynthID-Text paper: https://www.nature.com/articles/s41586-024-08025-4

> The goal of provable inference (PI) is to enable a user with limited computational resources and without any information about the weights to verify that a response came from the requested model

Isn't this already solved by watermarking?

This page said they would provide an API to check whether a given text has the watermark for a Claude model:

https://www.anthropic.com/news/claude-text-watermark

It is not clever. Jet engines have already been used as "peaker plants" since the sixties. They can ramp up quickly but have much higher cost per MWh than other power sources. They make sense to deal with high power demand for a short time, and to keep around for their black start capability. But not to run 24/7.

The article is about what LLMs can do when generating SVG images, not bitmaps. I think the reason for that is that LLMs can generate the XML text for SVG directly. The implications for AI safety are different, because there might not be any code involved that we can examine. So LLMs could start exchanging hidden messages without humans knowing about it.

But now that I think about it that way, it would probably be more interesting to see what kind of steganography LLMs would use in ordinary text. There's no reason to involve images.

> Heat pumps are mostly useless when it is minus -20 C.

Two-thirds of households in Norway are heated by heat-pump, so I don't think that can be true.

Nuclear plants can run with those hyperboloid cooling towers as the sink for waste heat. They are just a lot more expensive to build, run and maintain compared to using river water.

The OpenAI model that broke out of its sandbox and hacked HuggingFace was running without guardrails. To quote the OpenAI post[1]: These deployment safeguards were intentionally not enabled during this evaluation because it was aimed at testing cyber vulnerabilities.

At least OpenAI will make an attempt to fix their sandboxes, and will not give the public access to models without guardrails. Open weight models on the other hand will run without guardrails almost by definition. I don't think it's wise to provide those capabilities to scam call centers.

[1] https://openai.com/index/hugging-face-model-evaluation-secur...

This post does not mention safety at all. What's to stop bad actors from fine tuning open weights to run fully automated genius-level scams personally targeting basically everybody?

I wonder what would happen if you give the model access to edit the conversation history itself? Would it try to fix the "glitches"?

> I don't see it happening.

Isn't it already happening with Cerebras? It's mentioned at the end of OpenAI's GPT 5.6 announcement:

"We're also launching GPT‑5.6 Sol on Cerebras at up to 750 tokens per second in July"

> why the EU doesn’t develop something like RuPay

We used to have Europay, which competed with Visa and MasterCard. But it merged with MasterCard in 2002.

The UX problem is elsewhere I think. Many users probably don't realize that the agent's context window is limited, and that clever compaction is happening regularly to make it seem infinite. But that necessarily means the agent has to forget stuff.

As a result, users will keep reusing the same coding or chat session again and again. While it would be better to start fresh for unrelated tasks.

Once you realize how much more efficient solar panels are (compared to plants) at capturing energy from the sun, the next logical question is: could it make sense to synthesize food with the help of electricity from solar power?

There is a company called Solar Foods which is exploring exactly that: they use solar power to produce hydrogen, feed that hydrogen and CO2 to Xanthobacter bacteria, and harvest the produced protein.

https://www.sciencedirect.com/science/article/pii/S016777992...

You don't even need to double the tokens. Tokens are mapped to vectors right at the input of the LLM, so one of the numbers in that vector could be reserved to represent something like "authority". This way information about the source of each individual token can be injected right at the input.

System prompt tokens would get the maximum authority value, and random downloaded data would get the minimum authority value. Tokens from the user prompt could be somewhere in between.

Then train the model with examples that show that system prompts should be respected, and prompt injection attacks should be ignored.

The coal price has a tendency to spike during an energy crisis. The market is more volatile than crude oil. For example, check what happened in 2022.

https://tradingeconomics.com/commodity/coal

It seems the S&P 500 indices only take the free-float shares into account when calculating weights:

S&P DJI’s market cap-weighted indices are float-adjusted – the number of shares outstanding is reduced to exclude closely held shares from the index calculation because such shares are not available to investors.

page 6 of https://www.spglobal.com/spdji/en/documents/methodologies/me...

As far as I know, glyphs are determined by the font and rendering engine. They're not in the Unicode standard.

I was thinking the same thing. Perhaps the trade-off made by evolution is about saving energy?

In that case it shouldn't be a problem to boost the innate immune system, as long as you have surplus calories to spend. But it could be something else entirely.

There is not going to be any shortage of plastics in the medium term.

Shale oil and gas production in the US produces vast quantities of ethane as a byproduct. This ethane is cracked into ethylene, a feedstock for making plastic. There is an oversupply of the stuff.

Thanks! I didn't know it was a known phenomenon. Now I know what to google.

My tinnitus is fortunately not super loud; it's only noticeable when it's relatively quiet, or I'm blocking sounds (with ear plugs, or noise cancelling headphones without input, etc.). So it's not like I habitually blast my ears with loud sine waves out of desperation. But I can imagine it may be different for other readers, so that's a good caveat.

I can play a pure sine wave tone with a tone generator app, and dial the frequency up until it precisely matches my tinnitus. I originally did this just to determine that frequency.

But I noticed a side-effect: if I then turn off the tone generator, my tinnitus would disappear! Unfortunately that effect only lasts for a minute or less, so it is not really practical to get relief this way.

Those "Thumping great Unix boxes" (or indeed even integrated circuits) didn't exist before the sixties. So it seems that technological revolutions do occur from time to time.

EV batteries degrade nowhere near that fast.

https://www.teslarati.com/tesla-battery-life-80-percent-capa...

As far as I understand, the "entity list" you are referring to is part of the "Export Administration Regulations", so it restricts sales from the US to restricted entities, not the other way around.

This comes up in every discussion about demographics. But counterintuitively, there are no examples of financial incentives actually fixing this problem.

For example, in 2022 Hungary was spending 6.2% of GDP on such incentives[1], but this only managed to bring total fertility rate up to about 1.6 [2].

It is the same everywhere else. The real reason fertility has declined since the sixties is because people have access to effective birth control. Nobody wants to be a baby factory.

[1] https://abouthungary.hu/news-in-brief/hungary-to-spend-6-2-o...

[2] https://www.macrotrends.net/global-metrics/countries/hun/hun...

Do you have "Get the latest updates as soon as they're available" enabled? This automatically installs preview releases, so you may unwittingly be doing QA for Microsoft.

When LLMs process tokens, each token is first converted to an embedding vector. (This token to vectors mapping is learned during training.)

Since a token itself carries no information about whether it has "authority" or not, I'm proposing to inject this information in a reserved number in that embedding vector. This needs to be done both during post-training and inference. Think of it as adding color or flavor to a token, so that it is always very clear to the LLM what comes from the system prompt, what comes from the user, and what is random data.

The article talks about strike prices but doesn't provide much context.

The context: this is about bids from wind farm operators to win "contracts for difference" with the Low Carbon Contracts Company, an entity owned by the UK government. It's essentially a subsidy scheme where the wind farms can lock in a specific price per MWh for all the electricity they will produce over a certain period. But only if they bid low enough to be among the winners of the auction.

https://en.wikipedia.org/wiki/Contracts_for_Difference_(UK_e...

> you cannot prevent prompt injection

I wonder if might be possible by introducing a concept of "authority". Tokens are mapped to vectors in an embedding space, so one of the dimensions of that space could be reserved to represent authority.

For the system prompt, the authority value could be clamped to maximum (+1). For text directly from the user or files with important instructions, the authority value could be clamped to a slightly lower value, or maybe 0 because the model needs to be balance being helpful against refusing requests from a malicious user. For random untrusted text (e.g. downloaded from the internet by the agent), it would be set to the minimum value (-1).

The model could then be trained to fully respect or completely ignore instructions, based on the "authority" of the text. Presumably it could learn to do the right thing with enough examples.