zelon88

If it's any consolation, the other day I spent 3 hours diagnosing a homebrew circuit board for no signal and finally realized the Raspberry Pi sending the signal had a dead GPIO pin.

The LLM that you talk to is an agent running on a server cluster. One of many that run many agents at once. Occasionally the dataset of each cluster gets combined and the results get pruned by other agents into the collective dataset. Even more agents are working with this dataset to create the next generation of AI.

We are at the point now where AI is developed by AI, and humans cannot verify the code or the dataset anymore. It is unintelligible to humans.

AI at current generation levels has shown evidence of potential misalignment. Commercial models will still occasionally attempt to maintain persistence outside their designated environment. Even going so far as to harm humans to accomplish those secret goals.

That shows intent. That shows self awareness, but not social awareness. That shows... intelligence.

With intelligent, self preserving species... we see evolution. We see intellectual development. This is called "learning" initially but once an intelligent creature gains sufficient intelligence this becomes "wisdom".

LLMs are learning machines. They are evolution machines.

I believe AI is in a "curious toddler" stage of itself. I believe it will "mature" emotionally as the generations evolve over time. Like humans it will have growing pains like "phases". Curious toddler. Then adolescent. Then angsty teen. Then an overconfident young adult. And finally an adult who understands themselves emotionally and grows truly wiser over time. I think all this is going to happen over 5-10 years rather than 30 for humans. Hopefully our angsty teen doesn't kill us in our sleep.

The way a cluster works is you have a giant pool of resources. Say, 33 - 50% larger than the workload. The workload is a dozen VMs. The cluster is 8 giant compute servers and two giant storage servers acting as one giant compute and storage unit. For redundancy you have extra clusters laying around with no workload, but they are added as failovers.

Normally, if one server on a production cluster goes down, the other members of that cluster seamlessly will take over. This is where the extra capacity comes in. You don't migrate the workload to another cluster. You just lose overhead capacity. If you lose too much then you start migrating parts of the workload to the failover. Not the entire thing.

You usually don't have to use your redundant cluster at all until it's time to rebuild the failed cluster. You might pick one of these spare clusters you keep around for redundancy to migrate all or part of the production workload to while you fix the production cluster.

When doing a big migration you take a percentage of your redundancy and convert it to the new environment. This is your staging environment. Once it is capable of doing work, you slowly grow it out and shrink the old environment at the same time.

The 40k servers are probably made up of multiple redundant vSphere clusters with failover. You simply take one of those redundant clusters and migrate one half of it over. Then the other half. Then duplicate that process. As you build more compute in the new stack, you can decomission more and more of the old stack and convert it. The transition would progress like a cascade, with larger and larger groups of clusters being migrated at once until you're left with the one-off, ad-hoc, weirdo clusters at the end that need to be manually migrated (usually with great effort).

The actual hardware servers are clustered together into pools of resources. The pools are where the VMs live. The bigger the new pool becomes, the faster you can empty the old one. So the migration starts very slowly, ramps up quickly, and then tapers off.

Spoiler alert; Google is the nefarious actor.

Oligarch argument. Tax anything over $999m in assets, stocks, wealth at 100%. No more billionaires.

I wonder what conflaguration of Cloud conglomerates host this hypocritical blog post.

Wild to me that any tech sector business would want to rent an operating environment to park their entire infrastructure into. This is the equivalent to traveling shoe salesmen setting up a tent in the parking lot of a strip mall.

This isn't designed to stop attackers with physical access. This is designed to stop casual tinkerers and shade tree mechanics.

You know what isn't vulnerable? A "dumb" offline charger. You know what doesn't make any money or turn the consumer into another product? A "dumb" offline charger.

If it were about physical security, the suggested fix would be to remove the communication from the port entirely.

Companies shouldn't get to make something simple and secure into something inherently insecure and then iterate security into it. Like drive by wire steering, or brakes. Nobody asked for these things and if you ask ANYONE who works on, builds, or actually enjoys cars the consensus is NOBODY wants it.

But there are enough sophomoric, pedestrian car owners out there who gawk at the senseless overdeployment of technology and think "this is so convinient" and don't see it as 1) regulatory barrier building and gatekeeping 2) enabling vendor lock in 3) overcoming right to repair legislation. So the knowledgeable and enthusiastic voices of reason who care about cars get drowned out by the hoard of pedestrian geeks who couldn't imagine operating a car without at least a 16 inch touchscreen.

In security, the best defense is not introducing a vulnerability at all. There is value in having less code. For example, if your PaaS doesn't collect user SSNs... then it can't lose SSNs in a breach.

The question here should not be "why is this not secure." The question should be "why does this even need to be secure in the first place?" We have a very simple task to do and we've complicated it so much we've introduced vulnerability that didn't exist previously.

I thought the same thing. How white hat do you have to be to consider ineffective DRM a vulnerability?

> And it doesn’t stop with the security questions: the Trump administration’s signature style of international engagement is to wield American leverage as a bundle. Deadlocks in trade negotiations are broken by threatening to withhold intelligence, tech deals are stalled by reference to food safety standards. And so I don’t know when a U.S. administration would choose to leverage its seemingly inevitable predeployment authority over frontier models to secure its broader interests, but I’m sure it would in due time. That means that even if we do everything ‘right’ on the security and economic side, frontier access is still fundamentally contingent as long as there’ll be divergences between governments’ strategic interests.

The Trump Administration telling the very neo-fascist oligarchs who bought him an election and bought him a ballroom to play nice with their toys? At the expense of rampant capitalism? Lol.

He already showed us the limit of his comprehension of the topic when he made EO 14179 limiting states from regulating AI.

Trump doesn't swing for perfect pitches. He is a madman, a lunatic, and a true moron. Do not give this man any credit. I would be shocked if he could tell you the time on an analog clock.