Whether Anthropic was trying to hype about Mythos / Fable or not (and FYI, it is a pretty big leap forward), they absolutely did not want to get public access shut down. The US government very much seems to want to have exclusive access to it for now.

Also, to clarify the "jailbreak": They took open source projects that had known vulnerabilities, as well as deliberately introducing vulnerabilities into some other projects, then asked Fable to fix them, and then asked for test scripts to demonstrate that the exploits could no longer be exploited - the implication being that they could then use those exploits against unpatched systems. But what's the logic here? The challenge isn't "how to write exploits against known bugs", any model can do that. The challenge is finding the bugs - something Mythos / Fable has proven better than previous models at. Even if Fable refused to write said test scripts, it would automatically downgrade to Opus 4.8, and then *Opus* would have written those test scripts. Or any other model out there could do it, including free open source ones that can be safety-abliterated at will.

Is only half of women being able to read at all really the counterargument you were trying to make here?

And we're just talking the most basic aspect of an educational system. Kids today learn to read early in grade school.

multiple times. A database management utility needs to be installed on junior admin's machine, each time team staff rolls over.The app needs a recent JRE. Same for overseas end user who needs it for an integration. They either don't have Java at all or have an older version, invariably. Even people who should be responsible don't know how to upgrade Java, and maybe they have to go through a bureaucracy to do so. It is messy.

Office 365 (Excel) on a Mac. I had the Claude add-in in the ribbon. Today I discovered a Copilot add-in has been inserted to its immediate left. (I renewed my subscription but was unable to easily find a way to cancel and renew without Copilot like last year.)

When I decrease window width first Claude starts to disappear and Copilot icon becomes mini-sized, but Claude is not shown below it. Add-Ins ribbon button shows My Add-Ins with only Claude in it, Copilot not displayed.

In Preferences, select Copilot > Click the disable checkbox. Copilot ribbon icon dims and presumably will disappear on app restart according to MS.

p.s. I should mention the most popular related topic, what are called "AI hallucinations". It is kind of like a primitive brain, that grasps for concepts and then believes they are real, like citing a research paper that doesn't exist with a made-up title. Also things can creep into its "mind", a popular anecdote is telling an image drawing AI system "draw a room without any elephants in it". You will often get a picture of a room with many cute elephant images worked into the corners, in the drapes, the rug, on a shelf, in a picture frame on the wall. So yeah, it can be exciting but also a bit much!

Is there anything in this world you can say that about? Short answer: AI can be really powerful for some things but has some glaring weaknesses too. It is not a mature discipline. It has a limited reasoning capability and is heavily dependent on the amount (cost) of processing power thrown at it. It can make bone-headed mistakes unintentionally like forget things that scrolled past out of its context window, fail basic arithmetic, be drawn into conflicts with hidden directives from the vendor, lie when stressed, skip Excel rows to save time or sprint past chapters because it is tedious (happened to me), etc. Takes a ton of baby-sitting. But, if you understand how to handle it, you can get some impressive results. The problem is that people who don't know how to handle it get results that **look** good but have subtle stupid mistakes baked in, plus they stop exercising their brains which then atrophy. Fun times! :)

ED: EEG, not fMRI.

And again, that's not to imply that they have any particular "mastery" in this specific case. Obviously, if they just typed "write the essay for me" into ChatGPT and submitted it without reading it, then they're not going to have learned much of anything from that. The question is, however, what did they do with their time instead? Because their brain was learning that instead.

https://webs.uab.cat/saramarti...

While the author raises some good points, there are also problems. AI is apparently a major way to cheat but as a recent innovation it seems to me the lack of a rigorous education with proficiency testing during COVID when these students should have been gaining skills is more likely an issue. I wonder if a lowered attention span learned from addictive social media and a general increase in attention needed to digest more disparate pieces of information these days (whether news, entertainment, or whatever) may be erasons that students lack skills normally required at college level and have a lack of attention span. The above article suggests that rather than not being able to read, students are not willing to put in the effort to digest difficult topics. It might be due to not being native English speakers in this case so I'd say more testing is needed.

The correllary to "use it or lose it" is that the brain isn't just going idle, it's refocusing its efforts on other things that you are "using" instead.

The average person today could hardly identify all the wild edible plants in their area, change a horseshoe, or build a proper barn, like their ancestors hundreds of years ago could.

By contrast, their ancestors hundreds of years ago probably couldn't read.

Brains don't just go idle; they just refocus on different things. A wealthy Victorian often pursued a life of a polymath, seeking varied intellectual pursuits and sometimes making great discoveries, but they could probably scarcely tell you how to mend a shoe or even change a nappy - that was their servants job.

Also, it's quite the spin to present low MRI activity as "reduced function". It's commonly literally the opposite. If you present a novice with a task they're not used to, and an expert with the same task, the expert will tend to show much less activity than the novice, as the novice has to think harder to accomplish it, whereas it's become rote for the expert. Low activation on a task is commonly a sign of cognitive efficiency.

Yes, you're the second person to point out the typo, which would have been edited within seconds of me clicking "submit" if we had an edit button.

And also an inflationary currency is a very bad thing.

For one, it screws over debtholders (aka, most people), as the value of the debt owed grows instead of declining. If you work for a company making lumber, and a given board sells for $10 now, but 20 years from now sells for $5, and you sell the same number of boards per unit labour with the same relative margin, then all else being equal, your salary must be half in 20 years what it is today. But that mortgage that you took out today for $200k is still $200k (adjusted for interest and payment of principal). Which you have to pay on with a salary that is half what it is in dollars. That person is totally screwed.

Secondly, it discourages spending. The more you delay purchases, the more you'll be able to buy in the future. So everyone is encouraged to not spend. Which screws over your economy. It also screws over your tax base when taxation is based on taxing spending. Meaning you have to raise your spending-related taxes, which further discourages spending.

Third, it worsens wealth inequality. If you're living paycheck to paycheck, you have no savings. If you're a billionaire, you have a lot of savings. The billionaires see their assets grow and grow, and it comes at the cost of the working class. Also, said billionaires are encouraged to keep their money as cash rather than investment, which further ruins your economy.

This is no way to run an economy.

Another thing you REALLY don't want in an economy is instability. Aka, Bitcoin's fundamental nature, because it has no fundamentals and no attempt at monetary policy. Economies are fundamentally unstable. If you do not regulate them, they swing wildly. The faster an economy moves - and economies keep moving faster as technology advances, it no longer takes half a year for goods to arrive on a sailing ship or weeks to communicate with the other side of your country - the more unstable it is. This is terrible for both individuals and businesses.

What you want in an economiy is:

* Stability
* Low and steady inflation - a couple percent, to encourage spending, slowly devalue debts, and encourage investment. Too much is bad. Too little is bad.

Yet, silver is valuable as both a monetary metal and industrial metal.

Unlike bitcoin. You summed it up nicely right there.

Right now, I want to buy tin-silver alloy, to use in casting. But the prices of both tin and silver are insanely expensive right now. So I'm not buying now. If the prices go down? I'll buy. If they go down a lot? I'll buy a lot, perhaps even to sell on my sales website. And this buying is to *use* it. Which takes it off the market. There are fundamentals behind silver.

There are no fundamentals behind bitcoin.

401ks have fundamentals behind them (comprised of companies that make products and services that people want to buy, generally as repeat-buys)

Governments have fundamentals (the ability to levy taxes, backed by the full force of the courts, the police, and ultimately, the military)

Bitcoin has no fundamentals. It's a collectible. Its value is based purely how much people want that collectible. The only reason, as was stated, that people were buying it was as a lottery ticket. But there is no reason to "own" it beyond that. It's not generating dividends or doing stock buybacks based on profits. It's just there for those who want to collect it. And its value depends on how much people want to collect it.

(Arguably its greatest power is that its holders stand to lose so much if regulation goes against them that they tend to be very politically active, with large donations to pro-crypto candidates)

Eh... I think the "used for all kind of dangerous things" started with GPT-3. I used GPT-2 a good bit back in the day, and it couldn't pass as a person, let alone reason out as a cyberattack.

Since it wasn't instruction fine-tuned, you couldn't trust that GPT-2 would ever do what you wanted. The best you could do is "cute it" with leading text and hoping that it doesn't go off on a tangent. That said, you could kinda use it for things with decent reliability. For example, you could write a long paragraph, then write: "Or, to sum up everything above in one sentence" - and then (hopefully) get a one-sentence summary.

It could sort of try to write recipes ("Here is a recipe for [X]. INGREDIENTS:"), but far from usable (I remember that one contained vermiculite ;) ). You could write plays by starting out a play format, though the plotting was fairly incoherent. You could set it up to write a joke and it'd end up writing long anti-jokes where the punch-line never actually comes (in one, I set it up with "A nun, a priest, and a rabbi walk into a bar...", and it ended up with the bartender saying something antisemitic, everyone getting uncomfortable, and leaving, and priest turning to camera and talking about how antisemitism is bad). Etc.

It was this tantalizing glimpse of what was to come - not YET useful, but hinting at what could come if the models improved sufficiently. I don't think many people expected how quickly it would improve.

When GPT-3 came out (but not yet 3.5, aka ChatGPT), the writing quality was clearly dramatically improved. People were writing articles about GPT-3 using GPT-3, and only copyediting to piece the (small) chunks together. For me the big "Nothing Will Be The Same Again" moment was the "TheGentleMetre" scandal on Reddit, where it was discovered that someone had connected GPT-3 to their Reddit account... but it took like a month for them to get caught, and even then only because they were posting 24/7 at a regular pacing. And it was the fact that they weren't just writing arbitrary nonsense, but that there was logic to what they were writing. For example, one person asked something like "Would you rather have a psychopath or a Nazi as a roomate?" and it made a reasoned argument about how with a psychopath, there's something mentally broken with them, whereas with the Nazi, it's an ideology and there's still the possibility of changing that ideology. Back in that era, the notion that an AI model could do this was pretty staggering.

I mean, it is a pretty impressive model. And not just from the benchmarks. Relentlessly proactive and resourceful. It's a shame it's cut off.