Notices by zenkat (zenkat@sfba.social)

RE: https://mastodon.social/@jockr/117061713378575227

Remember that the story of your future is stories like this one, happening ever closer and closer to your home, until finally the disaster is upon you.

It's their story, it's your story, it's my story. It's ALL of our stories.

Unless we wake the fuck up and do something about it.

#climateCrisis

RE: https://c.im/@psoheil/116919484770655155

LOL the entire "debate" on whether we live in a simulation is ridden with so many internal contractions and unscientific woo. It is unprovable garbage that belongs at the bottom of your philosophy professor's trash bin.

The only reason anyone pays it any attention is that the tech billionaire class has realized you can buy unemployed philosophers for pennies on the dollar to fob off pet theories on a bunch of rubes.

The end game, of course, is convincing the masses that #AI has become the godhead, so that they can mandate its worship and permanently cement their power in a techno-theocracy (ref: Peter Thiel).

The funniest thing about the #goose presentation is how completely it misses the whole point of the fable.

In the original story, the goose has been providing a steady stream of golden eggs. But the farmer gets greedy, and decides that if he cuts open the goose, he can get ALL the golden eggs inside the goose at once. But after killing the goose, he finds no eggs inside, and is left with nothing.

This is a story about the financial class, #AI, and the #Internet economy.

@elduvelle @artemis Great analysis. As a software engineer and a biologist, I think your intuition is spot-on.

The biggest difference between LLMs and biological intelligence is that LLMs *don't learn from experience*, at least not immediately in the moment like we do. They have distinct "training" and "inference" modes. When you are interacting with Claude, you are in "inference" mode. It gives likely answers based on its training, but *it's not learning anything new".

The model has no memory of is interactions with you. Literally! All of the persistence you see in a session is really just a fancy trick -- they just feed your entire chat session back into the LLM, and ask "what is the next most likely token"? Again, and again, and again. Every time, your *entire conversation* is fed back into the model, just to predict the next token.

@inthehands When the fuck did warrantless stops without probable cause become legal in this country? 🤬

(Don't answer, it's rhetorical, I know. But fuck.)

@skinnylatte Wait, you go places without a jacket? That's like leaving the house without your phone.

@WarnerCrocker But the only reason they care is that it's hurting the Republican "brand".

They don't care about the murders. They don't care about the jackbooted fascism. They don't care about the chaos.

They only care that they'll lose votes.

@futurebird I was at Molasses Books in Bushwick last night, taking with my cousin about TBL's latest book, and the design of online systems. A shy but brave young woman asked if she could listen in. I introduced myself and welcomed her in to the conversation.

We talked a bit about various things, twitter and theater and art and connection. She was bright abd curious, made great points and asked great questions, had a charming sweet smile.

Near the end of it conversation, as we were getting ready to roll to our reservation at Win Son, she said something strange.

"Right now I think that ChatGPT has become my best friend."

I stopped and looked at her. "Don't trust it. It's lying to you. It's only pretending to be your friend."

"So I should trust people instead?"

"Yes. I mean, don't be foolish, but trust people over machines."

And then it was time to go. But I still want to go back and find her, give her a big hug, sit her down over a coffee, and tell her ...

DON'T TRUST THE ROBOTS

#ai #llm

@grimalkina Great thread, but I especially want to +1 this point.

A lot of my success in running software teams comes from being very mindful about keeping people's psychological defense systems from kicking in.

As soon as people get on the defensive, they are no longer working together to find the objectively best engineering solution -- you are fighting over semantic territory. We're hard-wired for it, when egos are threatened they will fight to win.

But threatened minds are not open or curious. They aren't taking in new data, they are defending what they already know. Listening stops, synthesis stops, creativity stops. And team dynamics go to shit, too.

It's not easy to get folks to engage and debate without feeling threatened, but it can be done. Takes a good amount of finesse at first, but if you can get it established as a cultural norm it can make magic happen.

@kims (missing link?)

@evan Everyone needs to watch this.

@WarnerCrocker Because they are paid to be the Loyal Opposition. The point is to make people think there is no alternative.

@inthehands Or immigrants who came here from authoritarian states, and know exactly where this is going. I have friends who grew up in Ceaușescu's Romania. They've been terrified since 2020.

@kims If you stared at the ad for more than 5 seconds, then they've succeeded.

@MediaActivist Makes more sense when you realize that while the US may be one nation-state, we are actually eleven distinct nations, each with very different cultures and values, glommed into a single state.

Not all Americans embrace white supremacy. The Yankees and Midwesterners have always detested slavery. The Deep South has always celebrated it. We fought a civil war over it, for crissakes.

Also important to remember that a key strategy of Russia psyops has been to inflame the internal divisions of their enemies to cause them to fracture and collapse. Now that their asset is in the White House, it's no surprise they are marching us to civil war. This has always been the plan.

https://en.wikipedia.org/wiki/American_Nations

@glyph I run a large software team. I find LLMs useful for summarizing related docs that are mentioned in designs and project plans I'm reviewing.

eg, There's a service I've never heard of that gets mentioned in passing. To understand the passage I need to know basically what it does, but exact details do not matter. Approximate knowledge is enough to get me unblocked.

Previously I would have needed to follow a link, skim a design, summarize in my head, jump back to the doc I care about.

Now I just ask for a summary of what it does. Instantly unblocked, stay in the flow of the design I'm reviewing. And if it's slightly mangled, who cares? It's not the design I'm thinking about, and I would have just as likely misunderstood trying to dig through the docs myself.

@dngrs @glyph Also note that I'm talking about systems that use RAG and large context windows to feed results directly into the prompt for summarization. I'm my experience, LLMs are much less likely to hallucinate in those cases, because they aren't reliant on latent information stored during training.

eg, If you ask an LLM, "tell me about the Floozbor API", most will happily make up bullshit for you. But if you take search results for [Floozbor API], feed them into your context window, and then ask for a summary, they'll do a decent job, because the key information is there in the input tokens instead of being extracted from the model weights.

@dngrs @glyph State of the art LLM summarization is quite good, far better than you give it credit for.

"Approximate knowledge" is an apt descriptor. Did you ever watch Adventure Time? There was this great side character called "Approximate Cat", who had "approximate knowledge of many things". This gave him great powers, but also glaring weaknesses.

Having a good LLM by your side when working through a massive pile of code & documentation is like having a tame Approximate Cat companion. There are times when approximate knowledge is a great power. And there are times when you better be on top of the fucking details or you're going to get smashed. The trick is knowing which situation you're in.

@dngrs @glyph
Here's another interesting use case -- large scale code refactorings. Our code base is millions of lines, and scattered throughout it are thousands of calls to a legacy macro that is no longer compatible with our monitoring systems. The translation to the updated version is relatively mechanical, but too complex to be encoded as regex or AST transforms. Updating it everywhere will literally be months of boring shit-work that even a new grad will grumble about, so it sits unfinished.

But LLM-based tools are getting to the point where they can do these transforms. Search the code, feed the localized snippet plus an example fix to the LLM, and it generates the change. Automate that and it provides patches for the whole codebase.

Now you still need to review all those changes, run the tests, do evals to show it doesn't affect result quality, go through launch processes. But it's just automated away a huge amount of shitty grunt-work that literally nobody wanted to do.

@dngrs @glyph LOL, so eloquent. Internet dialog at it's best.

Tell me, would you give a thumbs-down if I had used a regex in that scenario? What's the diff except an ability to be slightly more expressive in the types of transforms you can do?