bohaska

Replying toWhen the LLM isn't the one who's wrong

I tested this, and if you push Grok 4.1 to explain its claim in more detail, it'll admit that a second treaty, signed by the same delegation on the same day, includes the opium ban:

Grok's response

The treaty you are probably thinking of (same year, different treaty)

There was a separate treaty signed in 1880 between the United States and China that did address opium, but it is not the Angell Treaty.

Name: Treaty for the Suppression of the Opium Trade (sometimes called the 1880 Opium Treaty or Chefoo Agreement supplementary protocol on opium)
Signed: 17 November 1880 (exactly the same day as the Angell Treaty)
Negotiated by: The same American mission (James B. Angell, John F.

... (read more)

bohaska2moQuick Take

Where did the phrase "you can just do things" originally come from? I see it's the name of a new book now, but it probably has been in the water supply for a few years beforehand.

Replying toDancing in a World of Horseradish

bohaska2mo

Dancing in a World of Horseradish

The reason The Residence isn't very popular is because of economics. A Residence flight is almost as expensive as a private charter jet.
...
Rich people should just rent low-end private jets, and poor people shouldn't buy anything more expensive than first class tickets.

I think this needs more clarification. On a per-unit basis, yes, but the upfront cost of booking a private charter jet is a lot higher since you usually can't book only 1 seat, you have to book the whole jet.

For example, The Residence costed around $20,000-$30,000 at release (it's now more like $5,000, but the essay seems to be written with the former in mind). For a mid-range flight like UAE-Paris, you could book a charter jet for ~$75,000 that seats 10 people, which is a lot cheaper than The Residence at release, but is a bit excessive for one person.

[Anthropic] A hacker used Claude Code to automate ransomware

bohaska

6mo

Anthropic post title: Detecting and countering misuse of AI: August 2025^[1]

Read the full report here. Below lines are from the Anthropic post, and have not been edited. Accompanying images are available at the original link.

We find that threat actors have adapted their operations to exploit AI’s most advanced capabilities. Specifically, our report shows:

Agentic AI has been weaponized. AI models are now being used to perform sophisticated cyberattacks, not just advise on how to carry them out.
AI has lowered the barriers to sophisticated cybercrime. Criminals with few technical skills are using AI to conduct complex operations, such as developing ransomware, that would previously have required years of training.
Cybercriminals and fraudsters have embedded AI

... (read 812 more words →)

Replying toBanning Said Achmiz (and broader thoughts on moderation)

bohaska6mo

Banning Said Achmiz (and broader thoughts on moderation)

Tangential feature request: allow people to embed other comments in posts natively. This article uses screenshots of LessWrong to display conversations, but this does not responsively size them for mobile users and makes it harder to copy-paste stuff from this post, which a native implementation could fix.

Replying toBanning Said Achmiz (and broader thoughts on moderation)

bohaska6mo

Banning Said Achmiz (and broader thoughts on moderation)

Is this decision generally considered final and not subject to appeal, or do you expect comments on here/arguments by Said/etc to affect the final outcome you decide on?

bohaska6moQuick Take

If we have already found features/vectors for hallucinations, then why haven't major AI companies tried shifting it downwards when deploying their AIs? Does reducing their strength decrease hallucinations? Is there a reason why using it in practice would not be helpful?

About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong

bohaska

7mo

Update (20th Sep 2025): Scale AI has revised their Humanity's Last Exam preprint in light of this evaluation, and conducted their own checks on the accuracy of HLE questions, finding an error rate of 18% instead:

We conducted another targeted peer review on a biology, chemistry, and health subset, as proposed by [47], and found an expert disagreement rate of approximately 18%. This level of expert disagreement is in line with what is observed in other challenging, expert-grade machine learning benchmarks and also observed in other similarly designed work; for example, [6] notes that disagreement among expert physicians is frequent on complex health topics

They also note:

To illustrate, if we were to adopt a

... (read 912 more words →)

210

bohaska9mo

This sounds like a whiteboard to me

bohaska9moQuick Take

Epoch AI's new evaluation for Gemini 2.5 Flash Preview is broken.

On their AI Benchmarking dashboard, the newest Gemini 2.5 Flash model is listed as having an accuracy of 4% ± 0.71% on GPQA Diamond, when Google's official announcement lists it at over 80%, and when GPQA is a multiple-choice test with 4 options:

It's because of formatting issues. Helpfully, Epoch provides the logs from the evaluation, and the model just simply hasn't been responding in the correct format.

For example, if you look at the first sample from the logs, the correct answer is listed as "B", but the model answered in Latex, $\boxed{B}$, so it was scored incorrect. There are plenty of other examples like this.

Replying toWhat Does It Mean to "Write Like You Talk"?

bohaska9mo

What Does It Mean to "Write Like You Talk"?

"Write like you talk" depends on which language you are talking about.

Take Arabic. Written Arabic and spoken Arabic has diverged enormously compared to written English and spoken English. Modern Standard Arabic (MSA) is the formal written language for books, newspapers, speeches etc. But no sane person speaks it. There are a lot of spoken dialects (like Egyptian, Levantine, Gulf Arabic, etc.). A speaker of different dialects may not understand other dialects or MSA, because all the vocabulary and grammar is different, which isn't usually the case in English.

Written and spoken English are similar to each other compared to most other languages.

Replying toAI Can't Write Good Fiction

bohaska9mo

AI Can't Write Good Fiction

I've tried using a different method for r1 to generate flash fiction: one sentence at a time. If a human writer wouldn't write out a flash fiction story in one message, then AI shouldn't, either. Here's a result:

She clocked in at 6AM, categorizing discards by residue: toothpaste-crusted wedding bands in Tier 4, melatonin vials from red-eyes in Tier 7, a child’s sock curled around her sister’s garnet earring (missing since the November her calls went unanswered).
Room 312’s newlyweds left forensic poetry—dental floss strung bedpost to minibar, aspirin dust tracing slammed door trajectories. She logged these under Domestic Erosion, Subcategory: Honeymoon Phase.
Room 214’s grid collapsed at the gel insoles—mint-green, bunioned, size 6 like

... (read more)

A helpful page to see and subscribe to all 31 Substack writers (out of 122 total) who were invited to LessOnline: https://lessonline2025invitedlist.substack.com/recommendations

Consider showering

bohaska

10mo

I think rationalists should consider taking more showers.

As Eliezer Yudkowsky once said, boredom makes us human. The childhoods of exceptional people often include excessive boredom as a trait that helped cultivate their genius:

A common theme in the biographies is that the area of study which would eventually give them fame came to them almost like a wild hallucination induced by overdosing on boredom. They would be overcome by an obsession arising from within.

Unfortunately, most people don't like boredom, and we now have little metal boxes and big metal boxes filled with bright displays that help distract us all the time, but there is still an effective way to induce boredom in a... (read more)

Made a small, quick website showing GPQA benchmark scores plotted against LLM inference cost, at https://ai-benchmark-price.glitch.me/. See how much you get for your buck:

Most benchmark data is from Epoch AI, except for those marked "not verified", which I got from the model developer. Pricing data is from OpenRouter.

All the LLMs on this graph which are on the Pareto frontier of performance vs price were released December 2024 or later...

People are better defined by smaller categories.

If someone is part of both a large category and a small category that usually don’t overlap, it’s likely that they are an outlier in the large category, not a representative member.

For example, if someone is both a rationalist and a Muslim, you shouldn’t expect them to be very similar to a normal Muslim but much more similar to a rationalist, and that it’s possible that they may not be very good at representing Muslims in general to a rationalist audience.

UBI would probably work better in Kenya compared to the US because when people are in extreme poverty, funds get used to meet pressing needs, but when these basic needs are met, extra funds are probably more likely to go into having more leisure time. People in Kenya generally have more low-hanging fruit to buy a motorcycle or to get a home renovation, but in the US, major life-changing things probably cost too much for UBI to cover and the income might be used on something like alcohol.

Is the Renaissance caused by the new elite class, the merchants, focusing more on pleasure and having fun compared to the lords, who focused more on status and power?

Bohaska's Shortform

bohaska

This is a special post for quick takes (aka "shortform"). Only the owner can create top-level comments.

High school advice

bohaska

bohaska, Firinn

What advice does the Lesswrong community have in general for high school students who stumbled on this community?

Looking for all types of responses, it's fine whether your response is focused on achieving something in the AI/effective altruism/rationality community, or whether it is focused on a more general audience.

LESSWRONG
LW

LESSWRONG
LW

About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong

[Anthropic] A hacker used Claude Code to automate ransomware

Consider showering

High school advice

bohaska

[Anthropic] A hacker used Claude Code to automate ransomware

About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong

Consider showering

Bohaska's Shortform

High school advice

Why did Russia invade Ukraine?

bohaska

About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong

[Anthropic] A hacker used Claude Code to automate ransomware

Consider showering

High school advice

bohaska

[Anthropic] A hacker used Claude Code to automate ransomware

About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong

Consider showering

Bohaska's Shortform

High school advice

Why did Russia invade Ukraine?

The treaty you are probably thinking of (same year, different treaty)