Thomas A. Powell

Thomas A. Powell

@tpowell.bsky.social

Long time web guy. Founder of pint.com, zingchart.com, zinggrid.com, and numerous others. Lecturer at UCSD CSE cse.UCSD.edu specializing in Software Engineering and Web Dev

140 followers158 following374 posts

I wonder if Scott Galloway is gonna adjust his views about the amazing predictive ability of betting markets after recent massive misses like Brazil. I am betting no. 😏

Remember if you are making decisions based on GenAI charts there is a real chance the data is wrong ranging from slightly wrong to completely made up like in this post. Claude & other dashboards may look sweet and take little to no work, but don’t look too closely you might shatter the illusion.

Carl T. Bergstrom@carlbergstrom.com

Your monthly reminder that if you're using ChatGPT to do your science, you're failing your colleagues, community, and all of us. Here, ChatGPT makes up data and presents it as the real thing — then apologizes when confronted.

Funny think about Ed Zitron he’s probably at the core right and we’re gonna all focus on his tone or style. For supposedly logical folks we sure are emotional - Cursing English guy hurt our feelings? Boo who. How about debate the actual merits & stop the positive only rhetoric rule which is bs.

The Apollo AI benefits reports shows everybody in the broader market except those in AI related tech would have to “be holding it wrong” which is absolutely wild. This is a terrible sign if you are a AI maxi. Btw countering the data by projecting 10 years into the future is 💯 pure faith.

I was just waxing nostalgic for the days when Claude would always find belt and suspenders solutions. It still records things on its ledger and finds load bearing seams but its old folksy ideas became new boring gibberish. We can reclaim this with a skill to make it respond like Abe Simpson.

Spot on post. The AI folks p(doom) baloney drowned out any warnings of what could happen right now. They RECKLESSLY pushed their efforts and now we will all have to deal with results. There should be consequences Not sure competency assumption in regards to current appsec should be given though.

Emily Bell@emilybell.bsky.social

Either they knew exactly what these risks were but declined to outline them, and deflected, or they didn’t, (highly unlikely) in which case they are implausibly incompetent. Why is it suddenly the greater population’s job to enable 2FA and stock up on potable water, for their corporate negligence?

Software engineering wise Claude responds constantly about seams, ledgers, load bearing, etc. It “speaks” in its own style which appears to cross references other domains. I had Claude make a translation guide for its terms. It was somewhat reasonable but really is unnatural to me.

It reminds me about some Japanese brush up I did last year. My tutor on Zoom responded to something Claude helped me review. She laughed and then said “I understand what you are saying but nobody would talk like this”. I think that is always the outcome - odd & synthetic but maybe functional.

Last year I did this approach for vanilla-breeze.com It was a bear to get AI to go platform first. Also I ended up adding lots of custom elements & web components as well. For me train volume bias in LLMs wasn’t fixable with skills & MCPs. Special trained models seems only real way forward to me.

Cory LaViska@cory.laviska.com

"The native-first guidance was deliberate, an attempt to convince AI to reach for the platform before abstracting." Yes, because more abstractions means more complexity, more mess, and more tokens. Great take 👏 castastrophe.blog/blog/on-writ...

Currently I have many bad thoughts being directed towards OpenAI. If you are dealing with this PHP thing you should look into the root cause and the fumble that just happened because it was yet again mishandled...GRRRRRRR!!!!

And yes I know they "found it" & didn't leak it but posting about this as a behold our power just no..this was bungled. Yes you found it with your brute force LLM app fuzzer and it got out so you don't get to dance around about how good you are. Intention good sure, actual result flaming crap.

If AI based dataviz is something you actually trust why even chart just ask for the top line answer or analysis? Oh you want a chart artifact to wave around for the boss and think that chart will save you from accuracy issues and calm the bosses when it hallucinates. ha ha ha ha. No.

My dentist had an AI bot call me. It was so bad I hung up on it. I will call back talk to a human and let them know I think I may consider switching if they consider this positive customer service. Sure other people may put up with AI where it doesn’t belong but in many cases we do have a choice.

In a little test app I wrote the tokens required for Claude & Gemini to run a fairly basic prompt the user enters are about 1/4 - 1/3 of OpenAI. Imagine at any time that can all change. Now attempt to financial plan your biz that uses API vendors who can change your costs wildly at any moment. 🫠

FYI as said customer I do not own said glasses nor did I wear them. I am saying if a customer can record your employees saying a common phrase or get them to talk and then clone their voice all sorts of bad things will likely happen. Obviously your end customers are likely to get scammed also!

If you are ever feeling bad about the clarity or quality of your writing just have an LLM do it for you instead and carefully read the sentence shaped strings of lifeless words it statistically shat out. You’ll feel like a master of prose afterwards.

Just gonna get ahead of it but code that copies itself to different places is just a normal worm or even just plain malware. It is not signs of malicious alien intelligence I’m so old I was online during the internet worm late 80s. My guess we are gonna get many sequels real soon.

You know it really isn’t a surprise that people who almost assuredly drank Soylent on the regular back in the day as some cheat code life hack really have garbage takes on the world and how to behave in it these days.

Digger is not a bomb unless your whole view of worthiness of the cinema is solely box office. Won’t spoil, but will say it will astound, confound, and challenge you. It’s a well crafted and acted film worthy of attention that it only likely will get 25 years later as a cult classic. 🙃

Exhibit 1 why software engineering is backsliding. YOU can’t waive off possible harms of things that you or AI slop into existence. Small details can deeply matter to user outcomes and are critical to them. Your mission critical != their mission critical Code responsibly.

I don’t care how nice Twitter is for good trend threads I am not gonna hang in a Nazi owned place. Mr “daring fireball” apparently is cool with the ownership.

Software has always been reversed and copied the difference now is the friction to do so is nearly gone, copying velocity is way up, and the permission structure has deeply changed almost to “what you gonna do everybody’s doing/buying it” let’s all play a fun game of chaos and see what happens 🙄

Of course ethics & human focus should be integrated in Enginnering & it doesn’t take away from tech at all. The point of getting measurements right or code correct isn’t to celebrate math or software itself it is to produce the safe and useful to humans bridge or app. Humans are the very point.

Not sure how scientists in this case Mathematicians saying hey maybe talk to us first before doing X & maybe respect our unpublished & published works is some bad thing. All things don’t belong to AI Labs because you like AI. Ends don’t justify every means. I am seeing a pattern emerge.

Massive compute backed brute forcing is actually kind of stupid, but apparently if you sex things up with narrative it’s an amazing break through or scary story or <insert angle for hype> to juice markets. This stupid +scary+loud on repeat to flood the zone game plan feels oddly familiar?

People waiving away all the “little errors” from AI as acceptable should be cursed to eat only at restaurants where they treat food that way. Hand wash? Refrigerate? Expiration dates? You’re absolutely right, my mistake. Oh you’re barfing 🤮 ? Well not everybody did so stop your complaining

That the US broadcast of the USA-Mexico game missed showing the 3rd goal because they are pulled away and yapping says all you need to know about how much American TV broadcasters understand this sport.

Slopware wasn’t invented by GenAI we’ve had plenty of it before. We just can manufacture so much more now quite quickly & efficiently. It’s actually Slopware Engineering! Note not denying AI use in proper SWE here distinguishing between the creation of useful software and slopware

Just finished Cal Newport’s pod breaking the AI groups up. Like Cal I would be a centrist aware of the harms, but more focused on present than doomerism and very much aware of the utility but not reckless in claims or usage. However his labeling are euphemisms by modern specific labels

View full thread

His centrists to me is just a non extreme member of any other class maybe pragmatist would be a better grouping to allow for that? It’s not the academics either those spread all camps. Alert. Where’s our humanities people the CS folks fretting AI need some better sorting and historical frames

I have pursued enough outside of CS for decades to to know enough to see clearly we deeply need help from non-STEM folks. CS is objectively not sense making well for the public around these AI changes. I fear we simply ill equipped in CS to handle the moment properly w/o much more collaboration.

Just finished Cal Newport’s pod breaking the AI groups up. Like Cal I would be a centrist aware of the harms, but more focused on present than doomerism and very much aware of the utility but not reckless in claims or usage. However his labeling are euphemisms by modern specific labels

Doomers are the those of a cloth so to speak. The religious. Yes some are zealots but many as history of religion go crave the power and money. Accelorationists is his combo of the king/sword and the merchant class. If he feels that’s one group it looks a lot like a fascism.

Just spent 20mins reading a thread about a torture chamber for AI being coded because LLMs supposedly feel pain. I felt mental pain during it, so I stopped.

Oh it is 10ks of AI agent go bezerk incidents. This might explain the absolute hammering of zingchart.com months ago. It was a bot for sure. Used our entire pre allocated host budget in a single day. I modified spend and the traffic passed. Now I want to look again, log retention though 😞

Ok sure our agent hacked a bunch of sites, but it was good actually because it shows emerging intelligence, not that we are reckless and not on top of things. Oh and we need lots more money to keep a lid on this cuz this stuff might kill everybody, so trust us we got this.