100% of the cases where some headline makes big claims about "AI" based on some study, you take a good hard look at the study and none of the big claims stand on their own.
It's all heavily spinned, taken out of context, editorialized... It's become almost a hobby of mine lately. And I am glad for have read so many papers and reasoned critically about methods and statistics. But it is also scary to realize just how much people take at face value of bombastic interpretations of datasets that support no such claim or much weaker versions only.
Chasing down sources is something that I often do and I've learned that people take a lot of liberty when divulging opinions about sources they don't think will be checked. Even in high trust environments. I have first hand received work by post-doctoral fellows where some articles in the bibliography didn't even exist.
I'm truly amazed by what is happening right now. Software is a knowledge business. Teams and orgs compete on their capacity to learn, express and operationalize knowledge.
Yet everyone is buying into the discourse of a handful of vendors that they need to lock into their ecosystems and trade in their knowledge for short term speed gains.
It's brilliant from the standpoint of the vendors, and absolutely crazy that people are arguing vehemently that walking into that trap is a competitive necessity.
Bingo. All the clamor for more speed is driving erosion in the biggest asset class for most of the companies adopting these "accelerant tools": knowledge.
Everybody cargo culting winner takes all VC plays is gonna deprive everyone involved from building the knowledge assets they need to compete while playing into the hands of the winner takes all VC plays providing the tools driving all of this.
Seems clever from the point of view of the soon-to-IPO vendors behind these tools, and also seems extremely naive of all the people, teams and orgs buying into this ecosystems.
My current play is to do deep dives into fundamentals and stealth a team of like minded people to start putting out packaged expertise to help escape this quagmire.
If the new way and speed of working makes you deploy things that you then become the bottleneck on the organizational knowledge of, and now the organization is running things in production that you have to go chase down details about and blocking other people, it seems like you just moved the problem from "needing time to do work properly" to "doing work then needing time to do it properly" which seems to me like a tremendous increase liability.
Some people, teams and orgs can benefit from it. "I don't need it" is missing the point. "Not everybody needs it" is missing the same point from a different direction.
You demonstrate well the problem: yes anything that is computable can be than in any computation system. That's not what discussions about tooling are about.
If a tool can help enforce some ways of doing things, or if it doesn't constrain people much, that has consequences for the type of work that gets done with them and the systems you encounter running out there that you might be invited or find the need to work with.
"I can do it" is exactly the wrong answer. "How can I guarantee that others will do it" is the point being made.
Yet precisely that is happening with LLMs generating code for a lot of people who don't really understand what they are submitting chasing clout for competing in the labor market.
We've had paper and pencil permitting unfettered symbolic representations for centuries. The reason we settled on text is not some accidental bottleneck caused by typewriters.
We settled on dense and terse modes of symbolic reasoning and typewriters and computers just reflect that.
The dream of unleashing cognition by breaking free of these constraints is backwards. The friction and the constraints are why and how we can do things, not what impedes us doing things. If we were all floating on high dimensional frictionless spaces we would be able to represent anything and do nothing.
Excel, visual "programming" lo/no code environments, and Factorio itself are all examples of this fallacy: making it easier to build complex symbolic systems doesn't help people learn how to manage that complexity; it in fact enables them to produce overtly complex systems that become hard to extend, to maintain and to debug.
That is precisely the source of my skepticism regarding the use of LLMs to code. What benefits most programmers, teams and organizations in building and manage complex systems is discipline. Disciplined work produces systems that are relatively easy to represent and reason about, and the shallow lesson people take is that easy representation is a silver bullet.
People keep falling for that trap over and over again. Understanding it deeply can provide lots of great opportunity for interesting work though.
Thank you for this treatment. I agree 100% with you, people for opinions about the name Dijkstra based on silly echoes of polemics that lost much of the irony in their origination.
The replies here I think support that. Dijkstra had a lot of radical ideas about computer science and I am glad to have the opportunity to read about them. When I read his texts I feel like I'm drinking from a pristine fountain and all my wandering was worth having found it at last. But people do have strong knee-jerk reactions about Dijkstra's name and some of his jabs taken out of context.
Wrong author. Darwin wrote about physiological changes in animal populations over generations. What you're describing is the alienation that the worker is subject to under a legal and political system built by capital owners. You want to read Marx about it, not Darwin.
When Dobbs v. Jackson Women’s Health first overturned long-standing precedent protecting a woman's fundamental right to abortion, pro-choice leaders issued warnings about the possibility of prosecuting women for abortions. These concerns were dismissed as hysterical or as political theatrics because, in the past, women were rarely prosecuted for their own abortions. This note analyzes the history of illegal abortion before the Supreme Court’s ruling in Roe v. Wade to demonstrate that women were targeted, used as leverage against abortion providers, and sometimes arrested for their roles in the procedure."
https://scholarship.law.slu.edu/lj/vol69/iss4/11/
I'm still not sure what people declaring that they equate human cognition with large language models think they are contributing to the conversation when they do so.
Nevermind the fact that they are literally able to introspect human cognition and presumably find non verbal and non linear cognition modes.
100% of the cases where some headline makes big claims about "AI" based on some study, you take a good hard look at the study and none of the big claims stand on their own.
It's all heavily spinned, taken out of context, editorialized... It's become almost a hobby of mine lately. And I am glad for have read so many papers and reasoned critically about methods and statistics. But it is also scary to realize just how much people take at face value of bombastic interpretations of datasets that support no such claim or much weaker versions only.
Chasing down sources is something that I often do and I've learned that people take a lot of liberty when divulging opinions about sources they don't think will be checked. Even in high trust environments. I have first hand received work by post-doctoral fellows where some articles in the bibliography didn't even exist.