Snippings

xxx

Meta is bringing its hot new AI agent Muse to its smart glasses.

The move, announced Wednesday at its annual Connect conference in Menlo Park, California, gives Meta a new way to put Muse directly in front of users throughout the day.

Muse has become a fast-rising entrant in the increasingly competitive AI agent market: software that can take actions on a person’s behalf, rather than merely answer questions or generate text. Instead of opening an app or sitting down at a computer, people could eventually ask Muse for help through a pair of glasses already on their face.

With a connection to your Muse personal AI agent, Meta’s glasses “offer more ways than ever to help you manage daily life, whether you’re building healthy habits or navigating a busy calendar, all without reaching for your phone,” the company said in an announcement.

At Connect, CEO Mark Zuckerberg said the company will “start upgrading the AI experience on Meta glasses” in the coming weeks.

The announcement makes clear that the company sees its hardware business as important for the agent, which launched earlier this month and quickly rose to the top of the App Store charts, beating rivals like ChatGPT.

Meta’s broader hardware blitz at Connect also included Ray-Ban Meta Audio, a new audio-focused pair of glasses. It said it expects to offer more than 100 styles across Ray-Ban, Oakley, and Meta Glasses by the end of the year. It also announced Meta VR Glasses, which it said will offer a more immersive way to watch movies, play games, and work in virtual reality, and will go on sale next spring.

From: Meta Is Bringing Its Muse AI Agent to Its Smart Glasses – Business Insider.

xxx

A few years ago, back in 2019 in fact, I wrote that there was a revolution coming in financial services not because banks were getting AI but because customers were. I’m glad other people age. For example, Angela Strange, a partner at the noted venture capitalist Andreessen Horowitz, recently said that banks and insurance companies should be “TERRIFIED” by this development (her caps) because “inertia and information asymmetry will no longer be enough to keep customers”. Well, I’m not going to say I told you so, but… well, I did.

I was going to write something about this, even before I saw Angela’s remark, because of something I heard on the radio a few days ago. As you may know, Britain’s Air Traffic Control system keeps falling over. I don’t know why, but I assume it’s because the software was written by people and not AIs, but there you go. Anyway, I was listening to a discussion on the BBC concerning the disruption caused by these air traffic control failures. The specific issue raised was that people who had purchased travel insurance found out that they were not covered. A representative from the insurance industry said that people should choose travel insurance based on their specific needs for a trip and that they should check the policies on offer to see what is covered. At which point someone else mentioned that their inusrance policy document was 80 pages long. Yes: 80 pages. So if you wanted to compare, say, five different policies you would have to 400 pages, which is around two of the early Harry Potter books.

Next time I want to get some travel insurance, I will use AI. There is no point travel insurance companies advertising on TV, sending me junk mail, e-mailing me specific offers or launching targetted campaigns via business partners. I guarantee you I will pay no attention whatsoever so any of these.

Given that, how will a travel insurance company attract my business? What can it offer to my bot other than price? Since I have no idea, I asked an AI. Based on what it told me, here are a few suggeestions:

Machine-readable policies that agents can process without having to read the 80 pages, with the coverage terms published as structured data. It occurs to me that it might be useful to have relevant laws (such as the EU rules about compensation for delays and cancellations in an agent-friendly form too).

Proof of paying out. Verifiable claims data: how often they pay, how fast they pay and what they turn down. I think a lot of commerce is heading this way, frankly, moving us toward a reputation economy that rest of what can be checked, not just claimed. This, of course, links with digital identity which is point I will return to.

Room to negotiate. An API that lets the bot adjust cover to the trip, not choose from three fixed tiers, could be to the insurance companies advantage, because it can tailor its offer more specifically via a richer dialogue with an agent. As a person, it would drive me crazy going back and forth over the cover and the conditions, but this what agents were made for.

Automatic payouts. Cover that pays when a flight is delayed or cancelled, with no need to argue about exclusions. This is the sort of thing that smart contract and stablecoin fans have been talking about forever, and they have a point.

Sharing only what’s needed. Accepting credentials the bot can present, such as age or pre-existing conditions, instead of making you fill in forms. This again highlights the need for digital identity as a fundamental enabler.

What this boils down to it that insurers who want to win will be the ones who are thinking about to market to machines. There is a wholly new science of behavioral econonics for agents wating to be born here.

 

xxx

And the standard models aren’t just a bit cheaper than the frontier, they can be orders of magnitude less expensive. OpenAI’s GPT-6 Astra, for instance, costs $10 per million input tokens and $50 per million output tokens while Meta’s Muse Spark costs $1.25 per million input tokens and $4.25 per million output tokens.

From: What I Got Wrong About Meta’s Muse, And The Deeper Threat To Frontier AI.

xxx

xxx

Meta, it turns out, didn’t need frontier intelligence to build a really good AI app. Today’s ‘next-level-down’ models were good enough.

As the frontier labs spend hundreds of billions of dollars to develop the next levels of intelligence, the cheaper, standard models have become so competent that paying premiums for frontier intelligence is becoming harder to justify.

From: What I Got Wrong About Meta’s Muse, And The Deeper Threat To Frontier AI.

xxx

xxx

Crypto exchange Bitget suspects North Korean hackers may be behind a security breach that affected about $351.6 million in digital assets, citing preliminary evidence from an ongoing investigation.

Bitget CEO Gracy Chen said investigators identified internet protocol addresses linked to VPN services previously used by a North Korean hacking group. The pattern of the attack also resembled earlier operations attributed to the country, she said in a livestream on X earlier today.

Chen said the specific intrusion method used to enter Bitget’s systems remain under technical investigation.

Bitget detected unauthorized transfers from some wallets Thursday afternoon stateside, involving 19 transfers from parts of its hot and warm wallet infrastructure, while cold wallets remained secure, according to Chen.

Affected assets included ether, XRP, USDT, USDC, Avalanche and BNB across the Ethereum, XRP Ledger, Avalanche, BNB Smart Chain and Arbitrum networks. Earlier on-chain estimates put the outflows at about $183 million, but Bitget said those analyses had not captured activity across all affected blockchains.

From: Crypto platform Bitget suspects North Korea in $352 million hack.

xxx

xxx

Lloyds [RIC:RIC:LLBP.UL], NatWest (NWG.L), opens new tab and Barclays (BARC.L), opens new tab have now carried out two mortgage transactions using tokenised deposits, their industry association UK Finance told Reuters, while a trio of banks including HSBC (HSBA.L), opens new tab this week ran ​a person-to-person transaction simulating an online marketplace purchase.
The trials are part of the “Great British Tokenised Deposit” project run by UK ​Finance, after participating banks committed to the initiative through a pilot launched last year.
Banks say moving tokenised assets could be cheaper and more efficient than using existing systems, while the Bank of England has said it would rather banks innovate with tokenised deposits instead of stablecoins – ​a kind of crypto token pegged to the US dollar or other currencies.

From: UK banks make first interbank transactions using tokenised deposits | Reuters.

xxx

xxx

The report says verification should stay optional. I understand why – privacy and accessibility has to be built in properly. But optional puts the job straight back on the woman: know the badge exists, check for it, filter by it. The same research found women are already exhausted by that safety work.

From: Post | LinkedIn.

xxx

A simple way forward, as I have outlined before with respect to social media, is a traffic light system. Red means unverified, yellow means credentials valid, green means full disclosure. 

xxx

For example, attribute-based credentials could greenlight individuals holding verifiable credentials that attest to specific facts: age, citizenship, accredited investor status, absence from sanctions lists, without revealing the underlying data that generated those facts. A zero-knowledge proof can tell a counterparty “Yes, this person meets your requirement” without that counterparty knowing your name, income, or address.

From: SEC.gov | Looking for Change in Haystacks: Remarks before SIFMA’s Digital Assets Conference.

xxx

I was listening to a discussion on BBC Radio concerning the disruption caused by air traffic control failures. The specific issue raised was that people who had purchased travel insurance found out that they were not covered. A representative from the insurance industry said that people should choose travel insurance based on their specific needs for a trip and that they should check the policies on offer to see what is covered. At which point someone else mentioned that there inusrance policy document was 80 pages long. Yes: 80 pages. So if you wanted to compare, say, five different policies you would have to 400 pages, which is around two Harry Potter books.

Next time I want to get some travel insurance, I will use AI. There is no point travel insurance companies advertising on TV, sending me junk mail, e-mailing me specific offers or launching targetted campaigns via business partners. I guarantee you I will pay no attention whatsoever so any of these.

 

 

A group of four men dressed in black forced their way through the front door of a family home at 4 a.m., carrying at least a knife and an iron bar. The attackers bound the couple and their two young children, separating the father while placing the mother and kids under guard. For the next two hours, the attackers reportedly beat the father until he handed over his login details to transfer roughly €40,000 in cryptocurrency. Then they vanished. By the time firefighters found the family in shock around 7 a.m., the digital money was gone.