Ted

Clearly defining the problem is half the solution. When you say, "starved for feedback on how our work impacts the world", what does that feedback look like? Is it citations in other research, lower poverty rates, higher overall GDP, something else? Honest question.

Tech people have laptop stickers of laughing pizza slices and upside down smileys and inside is like the most gnarled async code holding up a distributed system for 10 million users

It also seems reasonable to expect AIs to eventually stop writing human-readable code and just write bits and bytes with the expectation that humans will never be expected to intervene. Should we consider legislation against doing this even if limited to certain contexts??

Writing code that is difficult to maintain and modify is irresponsible but unless the operator, me, _knows_ what unmaintainable code looks like, there's nothing to stop the AI from producing it. It is easy to imagine a future in which codebases are so complex that only AIs can understand them.

Well organized code also has fewer bugs, normally, because it is compartmentalized and thus easily tested. AI and AI-enabled coding tools are great at producing functioning code but never include tests unless explicitly told to do so. It's as if they have no concept of responsibility.

Massive effort has been put into making code maintainable. Maintainable code has low complexity, follows predefined patterns, is terse, etc. AI tools enable us to "fill the gaps" where coding ecosystems fall short. They allow us to shove square pegs into round holes thus increasing complexity.

1. LLM-generated code tries to run code from online software packages. Which is normal but
2. The packages don’t exist. Which would normally cause an error but
3. Nefarious people have made malware under the package names that LLMs make up most often. So
4. Now the LLM code points to malware.

How People Are Really Using Gen AI in 2025 Last year, HBR published a piece on how people are using gen AI. Much has happened over the past 12 months. We now have Custom GPTs—AI tailored for narrower sets of requirements. New kids are on the b...

In 2025 top 3 uses of GenAI are: therapy/companionship, organising life, and finding purpose. I think something has gone pretty badly wrong here.

hbr.org/2025/04/how-...

For the past year a colleague and I have been developing a tool for helping bid writers write bids for public tenders. We have not actively promoted it but the few who've tried it, love it. If you are interested in a demo, schedule me here calendar.app.google/Ec9gCP5rNMZG...

bidwin.ai

I'm on Spain's IRS (Agencia Tributaria) web site. I don't know if this is just a really bad translation or someone being facetious but apparently they have a "deprive you area". This is not a joke or fake. This is real!

The problem with climbing mountains is they never seen as big from the top as they did from the bottom.

You're probably right...

If that were true it would be called "reproduction-right", but it's not.

So, how, exactly does the copy get from the source (their web sites) and into the training set (a database)?

I hope this is good news. Seems like it should be.

Me (go)too

I'm actually honored that they made an appearance in my work. I've followed you for years but never had a close encounter before. Thanks!

??? seems fine to me… what am I missing?

Giraffes are still a thing in modern AI. In the attached transcript, I was trying to get ChatGPT to remind me of the word gobsmacked. I gave up after giraffe. @janelleshane.com

At work I want to be more valued for my presence than for my absence. In other words, I prefer that people value me being in their meeting rather than lamenting my absence… I know they sound like the same thing but they aren't!

RealPlayer circa 1997

American teens are flocking to a new video app

Done!

At my son's high school, students are not allowed to miss even a single day without a doctor's note. Doctors, however, tell us not to bring our children in unless they've had a fever for 3 days. Meanwhile, the kids attend class, spreading their virus and suffering.

History tells us that eventually this will become normalized (periods will be acceptable as decimals in Europe, maybe already is) and it makes me wonder how ultra correct LLMs will affect attitudes of the people that use them.

This is a very deep subject. I shall contemplate and post more later.

The fact is, much like with grammar, using a comma where we use a period in the US, is an "ideal" but in practice, nearly every European I work with is perfectly comfortable swapping periods and commas (and doing it the US way).

When thinking about bias, defining "correct" suddenly becomes very complex, but this, to me, is only a minor issue. The bigger issue is the potentially damaging effect on communities that don't subscribe to this hyper-correction.

One of the tasks is to judge the quality of the writing. For example, if I'm working on a European text, commas are used as decimals. If the output mistakenly uses periods (dots), I should flag the output as "incorrect".

I decided to sign up for a job helping to train an AI using RLHF. So far the experience has been… interesting!

I tell everyone I know: transitions are always the hardest part (as evidenced by this video - watch to the end)

Basically, if we're swapping writing code for writing prompts that write code, do we even care what that code looks like it it works and passes our end to end tests?

Will the traditional programmers job turn into writing tests?