rnxrx

I'm not sure we'll ever really be free of the GIGO (garbage in / garbage out) principle. Tools will get better and better, but can never be a substitute for a deep understanding of the thing we want to create.

Not to be glib, but is there any industry where management consultants have been shown to make a statistically significant difference either way?

I wonder if a part of the problem isn't just the misapplication of LLMs in the first place. As has been mentioned elsewhere, perhaps the agent's prompt should be to write code to accomplish as much of the task in as repeatable/verifiable/deterministic a way as possible. This would hopefully include validation of the agent's output as well. The overall goal would be to keep the LLM out of doing processing that could be more efficiently (and often correctly) handled programmatically.

Wouldn't this immediately put the American companies producing these models at a significant disadvantage? Just use an unmolested model hosted by a provider in Vancouver.

If anything, this measure seems like it would create a scenario where services hosted outside the US would become a lot more attractive relative to Trumped AI.

The general idea is to have the LLM maintain longer-term context/background by storing it in a format/structure that's akin to a standard Wiki. The result is (hopefully) a series of human-readable and editable documents that's developed and maintained by the agent.

There's great coverage of it at https://gist.github.com/karpathy/442a6bf555914893e9891c11519...

It's actually also now a base capability in the Hermes agent and has been really helpful for me, at least.

The arc of the technological universe is short, but it bends toward enshitification.

Awesome! I'll be sure to check them out.

The idea that the provenance of a given tool's code inherently pollutes the material it's used with seems kind of illogical. Wouldn't it follow from this premise that any code written using open source IDEs and debugged with open source debuggers and other tooling would itself then be considered copyleft? Are works written with LibreOffice not copyrightable?

There's obviously a huge issue with the legitimacy and ownership of training data being fed to LLMs. That seems like an issue between the owners of that IP and the people training the models and selling them as services more than the people using the tool. Isn't this just another flavor of SCO trying to extort money out of companies using Linux?

I love it when APL threads pop up here - really a novel approach to making computers do useful things. The videos from 1975 demonstrating it seem decades ahead of their time. I'm surprised APL hasn't had more of a resurgence given the recent ascendancy of data science.

- when all hope is lost in conversation, retreat into your self

I get a bunch of the others, but this one kind of confuses me. What's the opposite (better?) move? Logorrhea? Forcing a confrontation? Reciting bad poetry?

I suppose there's debate to be had about whether leaving a dead conversation is a retreat, but leaving the situation and pulling back a bit to think about the circumstances, one's mood, the people involved, etc seems like a good thing on balance.

It's not just code generation, either - more and more people in my own org are using Claude Code for infrastructure automation, devops, etc. Obviously some amount of code in there, but an absolute ton of tokens being consumed just dealing with Kubernetes work at scale.

Lots and lots of vendor financing during the dotcom era, and it ended up being a material part of those vendors' own difficulties. Especially when service providers were concerned (e.g. the huge crash in optical in particular).

Obviously it's not a perfect comparison, but you have to wonder how much of NVIDIA's income (for instance) is ultimately funded by its own money.

I've been using the NousResearch Hermes agent for the past couple of weeks and have to say it's been really good. I tend to hit multiple instances (and accounts)of Claude across three or four machines (between work and personal) and having a competent agent with constant state has been good for memorializing and organizing important info (directly into Obsidian, too), doing some amount of research and planning and it's also been helpful working out a lot of bugs with my burgeoning home automation setup. It's also been helpful dealing with management of several miscellaneous servers in the house, as it's definitely both faster and a better documenter than I am.

I have it running on a cheap VPS and it's fairly locked down. Especially with all of the self-reinforcement learning and skill development it's been improving its usefulness and, overall, I've been pretty pleased. Surprised even, if I'm being honest.

Gemini is featured just as prominently, and they've most recently been pushing their own model series (Composer).

This a great article - both interesting and potentially really useful to folks teaching- or learning- programming.

There's something I've never seen a good answer to: why is this being mandated in the OS vs requiring it for apps - or classes of apps? There's plenty of parental controls already available for browsers - after verifying the user's age on startup, why not add a header field that the browser inserts along with AgentID (for example) and call it a day?

This is probably fantastic news for the VPN providers. Lots of people who otherwise wouldn't have bothered are now likely incorporating VPN connectivity into their daily routine. This very obviously includes kids.

I also wouldn't be surprised if there were plenty of people only dimly aware of the idea of a VPN who are now sitting up and taking note.

Maybe there's an analogy to our long and short term memory - immediate stimuli is processed in the context deep patterns that have accreted over a lifetime. The effect of new information can absolutely challenge a lot of those patterns but to have that information reshape how we basically think takes a lot longer - more processing, more practice, etc.

In the case of the LLM that longer-term learning / fundamental structure is a proxy for the static weights produced by a finite training process, and that the ability to use tools and store new insights and facts is analogous to shorter-term memory and "shallow" learning.

Perhaps periodic fine-tuning has an analogy in sleep or even our time spent in contemplation or practice (..or even repetition) to truly "master" a new idea and incorporate it into our broader cognitive processing. We do an amazing job of doing this kind of thing on a continuous basis while the machines (at least at this point) perform this process in discrete steps.

If our own learning process is a curve then the LLM's is a step function trying to model it. Digital vs analog.

How about we just start with SCOTUS having transparent (and enforced) ethics and corruption policies?

I'm also increasingly aware that my own writing style and punctuation seem to line up with what might be associated with an AI, but some of the tells (em-dashes, spaces after periods, etc) seem like artifacts of when in history we learned to write.

I wonder how much crossover there would be between a trained text analysis model looking for Gen-X authors and another looking for LLM's.

The development of steam technology is a great metaphor. The basic understanding of steam as a thing that could yield some kind of mechanical force almost certainly predated even the Romans. That said, it was the synthesis of other technologies with these basic concepts that started yielding really interesting results.

Put another way, the advent of industrialized steam power wasn't so much about steam per se, but rather the intersection of a number of factors (steam itself obviously being an important one). This intersection became a lot more likely as the pace of innovation in general began accelerating with the Enlightenment and the ease with which this information could be collected and synthesized.

I suspect that the LLM itself may also prove to be less significant than the density of innovation and information of the world it's developed in. It's not a certainty that there's a killer app on the scale of mechanized steam, but the odds of such significant inventions arguably increase as the basics of modern AI become basic knowledge for more and more people.

There's a point in the article that mentions allowing the model to ask questions. I've found this to be especially helpful in avoiding the bad or incomplete assumptions that so often lead to lousy code and debugging.

The (occasionally) surprising part is that there are times where the generated clarifying questions actually spawn questions of my own. Making the process more interactive is sort of like a pseudo rubber duckie process: forcing yourself to specifically articulate ideas serves to solidify and improve them.

I think the progression of sentiment is basically the same. There were lots of folks pushing the agenda that connecting us all would somehow bring about the evolution of the human race by putting information at our fingertips that was eventually followed by concern about kids getting obsessed/porn-saturated.

The same cycle happened (is happening) with crypto and AI, just in more compressed timeframes. In both cases the initial period of optimism that transitioned into growing concerns about the negative effects on our societies.

The optimistic view would be that the cycle shortens so much that the negatives of a new technology are widely understood before that tech becomes widespread. Realistically, we'll just see the amorality and cynicism on display and still sweep it under the rug.

It's only a matter of time until private enterprises figure out they can monetize a lot of otherwise useless datasets by tagging them and selling (likely via a broker) to organizations building models.

The implications for valuation of 'legacy' businesses are potentially significant.

The other side of this argument is that we're constantly fed lots of extraneous information along with the actual interesting content. The point about listening to the storyteller is completely valid, but that story teller wasn't full of advertisements, links to other stories or entreaties to smash a like button.

To an extent we're becoming wired to skim content because that content has been so deeply interleaved with items that aren't just extraneous, they're not even from the storyteller. I'd suggest this capability is even a kind of survival skill, akin to not only being able to spot motion in a dense jungle but to also instinctively focus on certain kinds of motion.

I'm curious about why the performance gains mentioned were so substantial for Qwen vs Llama?

I had the same experience with Digital Ocean. Thankfully there were several other providers happy to take my money immediately.

FWIW, 100% that BGP itself doesn't *use* multicast, but it can *propagate* multicast routing information. It's certainly technically possible to support multicast on the Internet (..thus the invention of MBGP) but in practice has been a non-starter for a whole bunch of reasons.

It's kind of funny - a fair amount of the major network vendors' hardware (i.e. Cisco, Arista, Juniper, HPE) isn't that much better than what MikroTik has produced at a fraction of the cost. Having a better and faster processor is great, but I don't think it's going to move the needle very much.

This really highlights how much the OS on network hardware is actually the biggest barrier to entry to the larger market. It's arguably one of the market segments where open source has traditionally had the least amount of adoption. Things have certainly been changing in recent years certain use-cases (e.g. SONiC and similar for DC switching) but it remains true that the OS itself (and the associated supporting infrastructure) is actually what drives both adoption and stickiness, not the newest/biggest/fastest speeds and feeds.

It's been true for a while that if RouterOS could be enhanced and made more attractive (manageability, support, QA, feature roadmap, 3rd part ecosystem, etc) it would make MikroTik a major market disruptor.

This really does change the interaction with art. As a future expansion it might be neat to recognize images on camera that would make for interesting art (i.e. detection of people/animals or recognition of certain styles of composition) as well as being able to choose amongst different styles.

It seems sort of akin to some modern art that incorporated TV screens and video to make dynamic installations, like Nam June Paik.