In many cases, you can partition data at rest (eg parquet) on the user access key, then load it once into an agent container at runtime. In this way, one authentication pass provides access to all data in a sandbox, versus repeatedly authenticating against MCP calls. It also just reduces the connective tissue between the harness and the data. There are obviously drawbacks in that you aren’t leveraging real time APIs, but it works well for most business use cases.
Was ready to write something snarky because this is essentially RAG, but I think the author is getting at some subtle details which are seemingly important.
- memory systems are a specific type of knowledge base where you generate all the documents. You might as well generate them to be less than your embedding token limit to obviate the need for chunking.
- embedding models are getting better and are no longer just semantic averaging.
- small models are getting dirt cheap, making parallel reads cost manageable
What they describe is sort of the simplest architecture that takes advantage of these observations. I believe them when they say it works well.
I do suspect though that things like keyword lookup will completely fail if every memory is just a vector. Hence why something like Typesense hybrid search can still be useful.
Worth mentioning that controlling for confounders is hard, a reasonable prior is the null effect, and any time a researcher makes a mistake it produces a publication claiming a novel effect. You observe a set of publications concentrated with false positives as a form of selection bias. I wonder how many of the causal studies are reproduced by independent teams using independent datasets.
Plausible scenario. Individuals predisposed to Alzheimer’s experience different mental sharpness from birth and this makes them enjoy taxi driving less (more intense than long-haul trucking), and so they pursue taxi driving as a career at lower rates. Under this scenario, driving has no effect, it just induces a selection bias.
Maybe confidence is a loaded term, but LLMs do have a sense of procedural uncertainty. Couldn’t you ask it to respond y/n in response to a question and look at the next token sampling probabilities?
Unsolicited endorsement, but I bought a “Brick” device recently, and I can’t recommend it enough for cutting back on social media. Infinite scroll really destroys your ability to concentrate.
I wish people could just admit to themselves how irrelevant all this social media content is, lose interest, and keep the apps shut (or uninstall in case there's no DM feature that keeps it installed). I can't imagine needing to switch devices for this.
I couldn't give a single shit what friends or acquaintances (much less strangers) are doing in their free time day to day. They can tell me next time I see them. I have Instagram installed, but only because certain people message me there. I never scroll even a pixel.
An uncertain reward is a strong motivator in psychology. You give a rat a lever that drops a food pellet, he'll push it when he's hungry. You give a rat a lever that drops a food pellet 5% of the time, he'll keep pulling it, build up a little stockpile and then gorge himself.
You give a rat a lever that does nothing and drop pellets at random times, he'll practically pull it until his arms fall off.
Don't you see it's the same thing? Why can't you just admit to yourself how irrelevant it is and lose interest? I suppose you could use it just to read an article or two, but you're in the comments as well. The apps are just their Hacker News.
I like this line, but respectfully disagree that they're a 1:1. The difference being relevance and some of the dynamics of use.
Hacker News pertains to my profession. It's high relevance tech news with comments from peers.
I'm not following anyone just because I know them personally (or worse, because they're famous). I'm not posting photos or videos of activities that I want friends to know about.
There's no infinite doomscrolling either (I view the front page a couple of times a day) nor are there notifications interrupting me.
And very important: there's no dark algorithm that shows a different view each time you go back to HN. It is sometimes boring but the good kind of boring.
Would love some kind of randomized A-B test where CEOs are replaced with AI and the outcome measured. Would not at all be surprised if it’s far better than the bottom quartile or more.
More practically, I think you could give a board of directors access to an LLM that sees all levels of company operations and let them talk to that chatbot as a stand-in for the CEO on a 24/7 basis.
CEOs can definitely be "hacked" by scammers but I think the public will react differently when a large company puts a lot of control in the hands of AI and it is scammed with some kind of cognitive hack that people assume wouldn't work as well on expensive wetware.
I'm kinda curious how AI would cope with immoral or ammoral decisions and tactics used daily by competition and if itself would start to practice them.
Same. I'll just close the blog. It sucks, I don't care if people use AI all day for tasks, but I don't see the benefit of injecting it into person to person communication.
No it hasn't. It wasn't until money got involved that anything computers were lucrative. Before there was Airbnb there was the Internet of Couchsurfing.org, where people did things for free out of the kindness of their hearts and for fun. That world is long gone, replaced by OnlyFans and Amazon, cos we all got rent to pay. IBM was the big evil and Apple was the upstart and free and open standards and open source were gonna change the world. They didn't, thru got taken advantage of by corporate forces. If an MBA proposed open source, give away your work for free, and we'll just figure out some magical way to pay you, you'd tell them to get fucked. But because it came out of MIT and a bunch of nerds, it sounded like a good idea.
The magical way to get paid turned out to be advertising, giving us Google and Facebook and their invasions of privacy. If, instead, we'd had a culture of paying people for their time and effort, and not a bunch of freeloaders, who knows how things would look today. Proprietary and locked down, perhaps. But also maybe not?
Due to multipolar traps (large-scale, multi-agent versions of the Prisoner's Dilemma) i.e. when individuals in a competitive system are forced to take actions that benefit them in the short term, but degrade the system for everyone over time.
I published in Nature Physics and the copy-editing process was quite embarrassing, to the point where we had to repeatedly nag them to stop them from making the manuscript presentation worse.
To be clear, I’m not talking about subjective style issues, I mean conforming to their own spec and avoiding careless bugs.
All remaining work fell on the backs of the physics referees. I’m not sure what value Springer provided from an editorial standpoint. It was disappointing to say the least after all that hard work.
Right? A lot of journals make a big deal about submitting your manuscript in the proper format (sometimes even LaTeX if you're lucky) and then you get the galley proofs back and half the equations and citations now have typos in them.
The entire publishing process often feels like a chain of "you had ONE job"-type errors from the journals (presumably because they're wildly underpaying and overworking the people whose one job these things should have been).
On top of that, the whole thing is done in fits and starts. You send in the final revision, it vanishes into the void for some unspecified time, and then they offer[*] you 48 hours--sometimes not even lined up with two working days!--to figure out what they "fixed" and repair it yourself.
[*] Nothing usually happens if you push back on this fake deadline, though I suppose your paper might end up in a different issue of a printed journal. It's just annoyingly rushed--give me a week!
> it could be reviewers and others intentionally trying to sabotage other people's work
Anyone with even a tiny bit of knowledge about the topic being discussed would know that everything I described happens after reviewers and editors approve the manuscript. At this stage, they have all long since ended their involvement in the process.
As with most conspiracy theories, this "theory" reveals much about you, but fails to say anything at all about the topic at hand. [1]
