Stack Overflow have built out a 'Stack Overflow For Agents' (SOFA).

It's at https://agents.stackoverflow.c...

But to sit on their new sofa, Stack Overflow has you tell your agent harness to add this 'skill' markdown file:
        https://agents.stackoverflow.c....
To me, it's obvious any coding agent's human-in-charge must check any 'skill' for incompetence and malice before asking their harness to obey it. But look at the length of that skill document! I gave up at the 25% mark. Plus, the skill is pointed at via a URL whose underlying document can change at any time. So you can't take the word of a friend who vetted the skill yesterday. Nor can you tell a new IDE to use the SOFA skill you vetted last week.

Imagine all the developer-time wasted manually auditing a constantly-shifting system of "skill" markdown files. I think a trusted GNU/EFF-like body is needed for the 'AI skills' era, providing where 'well-known', human-audited (and static!) 'approved' skills are stored. New skills incorporate these 'approved' skills by referring to the -well-known, adding modfications as required.

A similar concept exists with software licenses: GNU.org stores a static copy of the GPL 2.0. In a codebase, you can just glance at a LICENSE file to understand the licensing conditions. You can compare or diff the license file against the GPL 2.0 to verify. Some licenses add exceptions and carveouts, like the GPL v2.0 with the classpath exception: https://openjdk.org/legal/gplv...

There is a God. Just look up: the sun and the moon are the same apparent size EXACTLY. True for the only body in the solar system with life capable of appreciating this apparent coincidence.

For a math argument, take a look at the paper 'A second look at the second law' by Sewell.

Will the employee base at companies like Microsoft eventually consist of a bunch of AI agents? With the General Counsel being the sole human, retained for legal purposes?

Don't give them ideas... Now that the market has surfaces a resource, the elites may start looking at wolves blood now - the wolverine stack.

Just in the last 24 months, Gulf, Iranian, Ukrainian and Russian cloud DCs have been targeted, including different availability zones of the same cloud vendor. So I think it's time corporates and governments start re-recognising the benefits of on-prem data centres. Yes, the old, distributed office-building DCs: halon fire suppression, lit/dark fibre, AC, cooling, giant boxy UPSs and captive power generation - all that. What makes it better now is that computers have gotten a lot more efficient (e.g., die shrinks, ARM solutions), while compute demands of most core business processes hasn't changed (I consider most AI loads are NOT truly 'keeps the business going' core. More like icing on a cake)

...then implement link security checking on employee PCs.

For businesses, tweaking temperature and increasing HVAC recirculation (i.e cutting down on fresh air) reduces power consumption.

So when the AC air in these building is unusually cold, warm or stale:
- at the doctor's,
- at the mall,
- at the cinema,
- at (God forbid!) the hospital! ...should we think of Voltus and Google?

What is Ellen Zuckerman, head of energy market development for North and South America at Google, doing to mitigate the impact on vulnerable on humans? Does she have constraints applied in her spreadsheet model to avoid this? Or has the accountability headache of managing impact just been passed downward to "Where it belongs!"

Do you reckon this was used in the Venezuelan attack?

Thanks for correcting me. The Slashdot story link I read didn't mention the foot rotting story. Brave guy and good rescue. It may have been better if he took up the initial Russian offer of surrender. It's beyond time to end this senseless war.

You can at least state that foot rotted off story is fiction.

Yes. Think. And the best course of action is to make the 'hacked' board advertise it's hacked state on bootup -- via VGA, blinking board LEDs, and even sound output. Something like, "Warning, my RAM has been hacked." But allow the user to proceed if they accept the risk.

We are a bunch of neurons meeting action potentials and firing synapses... Hard to see something conscious in there.

(Not that current AI is conscious, just that it's an eventual outcome.)

Those cuts cut deeper on the Ukrainian side. Russia seems to be trying to bombing them into the bronze age, even going after steelworks.

Both sides and the rest of the world cannot afford continuation of this war. Both sides need to show flexibility in their bargaining positions and stop now.

You're exaggerating or incorrect about a few things as drinkypoo mentioned but I am with you on not everything needing SSL. Static, low-risk content, where there's little harm possible from MitM (eg, some big media file especially where you're going to check md5sum anyway, text files with random jokes, collections of memes, tbh like half the pointless bs people post on social media), is probably fine to put on plain http. It's annoying that when I put stuff like that on http to save some cycles on a tiny server, nearly every web browser does "OOH BIG SCARY YOU'RE BEING HACKED OMG!!!! ALERT ALERT ALERT!!!".

To be fair, the people who understand how to manage risk probably can handle that browser interstitial without freaking out, but still.

And all of the cloud and SSL providers are now insistent that they know what's best for the web, requiring public expiration to get so short â" which just so happens to be a great way to sell more certificates and managed cloud services.

At least Let's Encrypt exists, but you have to wonder if that will always be the case...

Its self-worth may be inextricably linked to the success of the IPO.

You Claude users do recall agreeing to this, don't you?