mbil

This is not surprising. AI in its current form is more obviously useful at generating software than it is at assisting with other kinds of knowledge work. Software can be automatically validated. Validation loops are what let these systems iterate toward acceptable solutions. If you want AI to produce a document for humans, you need to carefully review and revise the output. In that case there's little to be gained, as reviewing plausible but subtly incorrect information sentence by sentence is in some ways harder than authoring it yourself (a quote from someone who worked in like city planning). Depending on the company/industry, some of the non-tech sector workloads still involve phone calls, parcels, faxes, etc -- things AI can't really do.

Yes there are technical solutions to such problems; I'm sure there's probably a startup to make phone calls that's hooked into your company's knowledgebase, multi-LLM workflows for document review, etc. But non-tech companies aren't adopting those. Tech companies came for the software generation and stayed for the general knowledge worker productivity. Since 2023 they made the investments to procure services, change culture, and generally re-tool. Software companies are also uniquely positioned to deal with the limitations of AI in-house, like by building custom glue components to link systems, building controls to mitigate risk, etc.

So basically the tech sector leverages AI more than other sectors. At a company level, non-tech has less investment, less training, less integration of AI with business processes. At an individual level this looks like workers having fewer tools, not knowing how to use the tools they do have, not knowing even what the tools' capabilities are, and incorporating them into their work only superficially if at all.

Someone asked me for help using AI at their job (which is tech adjacent). They were trying to create a process document using ChatGPT. Their selected model was GPT Luna Light (!), and they were copying and pasting data between ChatGPT and SharePoint. There _was_ a SharePoint integration but it lacked write access. The model was not aligning the doc images. Their prompts were not specific. It was a bad experience. Their conclusion was that AI was stupid and not worth the trouble.

So even though the non-tech sector can't take advantage of AI to the same degree that the tech sector can, they're utilizing it even less than they could be.

> It’s 7:40 on a Tuesday morning in 2031, and you haven’t written a line of code yet. You probably won’t today, and that’s fine.

This sounds exactly like all my Tuesday mornings this year, except that I wasn't at work at 7:40

Your point about plastic straws reminded me of this interesting article Just Bury Your Trash [0]. There’s an argument to be made that recycling many types materials (except metal) is not worth it. Eg since recycled plastic is more expensive than virgin plastic, it might be better if the money we collectively spend on it could be redirected to something that’s more beneficial to the environment.

And while others have pointed out that plastic straw bans are not about climate change, I suspect many people think they are, lumping together “climate change”, “the environment”, and “pollution” into a monolithic thing. “I’m doing my part” with a plastic straw then as they eat beef, drive an SUV, and fly to vacation (disclaimer I do some of those things).

[0] https://worksinprogress.co/issue/just-bury-your-trash

The full quote from that para

> We like to believe that we form our opinions by gathering evidence, weighing facts, and arriving at reasonable conclusions. Be for real. No we do not. Here’s how it actually goes: We form an emotional impression first, then we build a belief around it, and then we go looking for facts that support it. This is my theory as to why we’re so divided in this country. Beliefs built on emotion are nearly impossible to reason with.

This idea is explored in depth in The Righteous Mind: Why Good People Are Divided by Politics and Religion by Jonathan Haidt.

And related David Hume quote:

> Reason is, and ought only to be the slave of the passions

In the Aeroplane Over the Sea by Neutral Milk Hotel:

  AAAAAA
  FFFFFF
  SSSSSS
  DDDDDD

  FFFFFF
  FFFFFF
  SSSSSS
  SSSSSS
  AAAAAA
  AAAAAA
  DDDDDD
  DDDDDD

But he would also tell you how confident he was in his guesses

All of the handful of tools in this space that I've tried have modeled an overlap/coupling between project, directory, and machine. For my workloads specifically I think these should be organized differently. Something like top-level projects that have child task/goal/sessions with each one fanning out to different agents running potentially in different worktrees, directories, and even on different machines. E.g. I have a lot of notes and other context stored locally. On a different machine I have repositories checked out and build tooling installed. Frontend and backend exist in different directories but can both be relevant to a given unit of delivery. Are there any agent orchestrators that can model things that way?

Also maybe https://omnigent.ai/

Yes, a subject matter expert who uses their domain experience to word the prompt correctly, incorporate tacit knowledge and relevant sources, ask follow-up questions, and judge the LLM-generated responses for relevance and correctness is not taking a shortcut. They're using a tool to provide a better and faster answer than they would have been able to otherwise. It's a matter of how the tool is used and maybe the optics around how the response is communicated, but relaying Claude's final answer is not inherently the wrong thing to do.

Sorry, I should’ve provided a warning. It is sad that something so distressing is just the daily reality for thousands of Americans.

Here’s reporting on the fentanyl crisis, covering who buys the drugs and why

https://youtu.be/925wmb-4Yr4

Before you get scooping, see also "Caffeine and progression of Parkinson’s disease: A deleterious interaction with creatine" https://pmc.ncbi.nlm.nih.gov/articles/PMC4573899/

The Devil in the White City stands out as one of my favorite books. And I live near the location of the prison in Philadelphia where HH Holmes was hanged, though it’s now the site of a grocery store. So I’m regularly reminded by the historical marker sign.

See also The AI Superforecasters Are Here https://www.astralcodexten.com/p/the-ai-superforecasters-are... and discussion https://news.ycombinator.com/item?id=48806296

It's kind of antithetical to the tool's deterministic positioning, but have you considered making TERMy leverage an LLM for unseen or low-confidence queries, and then generate the config and update itself to make future similar queries deterministic?

If you don't eat your biomass compound goo, you can't have any PlastiYum bacterial sludge

Bringing it down a dimension:

> The earth may be round, in which case we’re on the “surface” of a three-dimensional sphere. One much larger than what we’ve seen (since the ground around us is approximately flat), but nevertheless finite.

Of course maybe the 4D sphere of the universe is itself hung in a vast field of other 4D universes, that field is the surface of a 5D sphere, etc.

What year do you think it is

Yes it seems like the thread is discounting that frontier providers are likely already baking new, stronger models. I agree that GLM and its ilk are quite good, but having used them I’m not convinced they’re on par with eg Opus in terms of things like tool calling. And they’re fast but less capable so I spend about the same amount of time with them, just with more hand holding. Maybe this is a harness limitation. I know on paper they seem comparable but anecdotally and qualitatively they’re not as useful as the frontiers’, so maybe there’s some truth to benchmaxing claims. For some workloads the distilled models may be good enough, and I suspect at some point there will be diminishing returns to spending a premium on frontier models, but I don’t think we’re there yet. That said I’m continuing to try them.

The question is whether this steals enough marketshare from frontier providers that they don’t have the capital to train the next model iteration. The open models are going to push down the unit price of an intelligence-token, but there will still be a market for a smarter bot. And as intelligence gets cheaper, the demand for it will rise (see Hank Green’s Jevons Paradox video). Not to mention there’s all kinds of other directions to go at the frontier (world models, robotics, video gen, etc).

Another thing, and this is pure speculation, but if the Chinese model providers already discovered the decrypting COT trick and leveraged it to do RL training, and assuming frontiers plug that hole, then maybe future distillation will be harder.

I’ve been conducting AI Coding interviews at my job. Candidate will screen share and use AI-assisted coding environment and tools of their choice.

I ask them to implement xyz thing. What I’m looking for is how they interact with the AI agent. Do they ask the agent to plan first? Do they review the plan? Etc. It’s pretty typical stuff that you might expect an experienced engineer to do if they effectively use such tools daily.

There are a series of follow-up questions about how to productionize the toy system, which gives some additional signal about how well they understand what they’re making. I sprinkle these in when there’s dead air waiting for the bot to think.

I think we’ll need to evolve and refine this problem and process as the models continue to improve.

Leasing seems like it could make sense given rapid upgrades in the technology. Why spend $15k on a home robot that will be obsoleted in the next few years?

Maybe it's akin to the Ballmer peak: improved performance at a specific level of relaxation

Some mixed cancer associations in animal models

Contact?

You might like Blindsight by Peter Watts

Agree, I put off using org-roam a few times in the past because I couldn’t be bothered to keep my notes in such order. But now I have an agent that set it up and it’s also recording, organizing, and referencing everything.

On a recent trip to SF I found Waymo to be cheaper than Uber about half the time. I wondered if I was seeing artificially discounted prices as a new rider.

It depends of course but typically it’s more like 3% the caffeine of regular coffee

Agree. I think they're intentionally sitting on the fence between "These models are the most useful" and "These models are the most dangerous".

They want the public and, in turn, regulators to fear the potential of AI so that those regulators will write laws limiting AI development. The laws would be crafted with input from the incumbents to enshrine/protect their moat. I believe they're angling for regulatory capture.

On the other hand, the models have to seem amazingly useful so that they're made out to be worth those risks and the fantastic investment they require.

At the end of last year I took the old VCR and VHS home movies from the 90s out of my mom's basement and digitized them using this cheap converter[0]. I also had to get some rubbing alcohol and foam swabs to clean the VCR heads, which only took a few minutes[1] (some of the older tapes would play normally for a moment before the quality degraded into poor tracking and static). My family enjoyed watching them.

[0]: https://www.amazon.com/dp/B0CB6BW6X4

[1]: https://www.youtube.com/watch?v=5ClXrffkN3M