Snippings

xxx

Anthropic also looked at how Claude solved simple math problems. The team found that the model seems to have developed its own internal strategies that are unlike those it will have seen in its training data. Ask Claude to add 36 and 59 and the model will go through a series of odd steps, including first adding a selection of approximate values (add 40ish and 60ish, add 57ish and 36ish). Towards the end of its process, it comes up with the value 92ish. Meanwhile, another sequence of steps focuses on the last digits, 6 and 9, and determines that the answer must end in a 5. Putting that together with 92ish gives the correct answer of 95.

And yet if you then ask Claude how it worked that out, it will say something like: “I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95.” In other words, it gives you a common approach found everywhere online rather than what it actually did. Yep! LLMs are weird. (And not to be trusted.)

The steps that Claude 3.5 Haiku used to solve a simple math problem were not what Anthropic expected—and they’re not the steps that Claude claimed it took either.
ANTHROPIC
This is clear evidence that large language models will give reasons for what they do that do not necessarily reflect what they actually did. But this is true for people too

From: Anthropic can now track the bizarre inner workings of a large language model | MIT Technology Review.

xxx

xxx

Anthropic used circuit tracing to watch its LLM Claude 3.5 Haiku carry out various tasks.

From: Anthropic can now track the bizarre inner workings of a large language model | MIT Technology Review.

xxx

xxx

the migration of readers from the web to AI summaries will likely continue. The internet has always favored ease. And using generative AI to find the most valuable parts of the web’s evergreen content — like recipes, personal finance, and history content — can be a better experience than poking through sites one by one. Along the way, these systems will likely rewrite the economics of the web, and perhaps reshape the internet itself.

From: As AI Takes His Readers, A Leading History Publisher Wonders What’s Next.

xxx

xxx

Meanwhile, constant demands from banks and mobile operators to comply with government-mandated “know your customer” regulations have been a boon for cyber fraudsters. Repeated requests for personal data to set up new ids are similarly conditioning Indians to hand over information to anyone who asks.

From: India is obsessed with giving its people “unique IDs”.

xxx

xxx

The Department of Animal Husbandry and Dairying is allotting unique ids to every cow and buffalo in India.

From: India is obsessed with giving its people “unique IDs”.

xxx

xxx

Amazon has unveiled an AI agent trained to independently perform actions, such as order and pay for things, within a web browser.

From: Amazon launches AI agent that can shop online.

xxx

xxx

UK startup thankyü has unveiled an app that lets customers tip hospitality staffers with a tap of their phone against a wearable tipping band.

From: Wearable UK tipping app launches.

xxx

xxx

The proportion of fraud originating on WhatsApp and Telegram rose steeply in the second half of 2024. Revolut notes that consumers are led to believe these platforms are secure, yet the number of cases stemming from Telegram jumped by a staggering 121% in H2, while Whatsapp cases rose by a similarly concerning 67%.

Despite this shift, Meta platforms (Facebook, Instagram, WhatsApp) collectively accounted for 54% of all scams reported to Revolut globally, marking the third consecutive reporting period where Meta has held this position.

From: Fraudsters turning to encrypted messaging platforms to scam users Revolut warns.

xxx

xxx

The proportion of fraud originating on WhatsApp and Telegram rose steeply in the second half of 2024. Despite this shift, Meta platforms (Facebook, Instagram, WhatsApp) collectively accounted for 54% of all scams reported to Revolut globally, marking the third consecutive reporting period where Meta has held this position.

From: Fraudsters turning to encrypted messaging platforms to scam users Revolut warns.

xxx

xxx

The UK’s reliance on US-dominated card networks is near-total, especially since the domestic Switch scheme was sold to an American company. This leaves open banking as the only viable alternative. But can open banking realistically replace established payment networks if geopolitical or economic tensions escalate?

From: Can open banking stand alone as the UK’s payment infrastructure?.

xxx

This is not only a UK discussion, it is a discussion across Europe, where the Commission has for years been searching what I could not resist calling “Le Third Scheme”.

 

xxx

No, the problem isn’t that Europe doesn’t have an alternative card scheme. The problem is that Europe doesn’t have an alternative to cards, which is why “Le Third Scheme” should be based on the things that Europe does have: Open banking, instant credit transfer, smartphones and payment institutions.

(With this in mind, I was pleased to see that the European Payment Initiative, the EPI, abandoned its plans for a card scheme — which I always thought sub-optimal — and decided to focus on an account-to-account instant payment solution (A2A) for all kinds of use cases, all through a wallet. There is an interesting synergy here with the European moves to develop a common digital identity service and euro-wallet infrastructure, but that’s a topic for another discussion.)

From: A Real Alternative To Cards In Europe.

xxx