Thomas A. Powell
banner
tpowell.bsky.social
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
Today I am absolutely right quite a bit. No I am not smart or correct I am using Claude too much.
November 16, 2025 at 12:15 AM
Stay off the main thread is kind of old school webdevs version of “get off my lawn” to the “kids.”
November 15, 2025 at 5:34 AM
Just tried an AI insta app builder. I had it build a vanilla JavaScript galaxian clone and it made me a React+typescript game. In my game a triangle could shoot a circle by a mountain of garbage code with the quality of late night gas station food. Sorry gas stations I was mean, but it sucked.
November 15, 2025 at 5:00 AM
Considering how many cyberpunk anime and manga focus on nature and people pushing back against corps and technology it totally tracks that Japan would lead the fight about OpenAi ripping artistic work.
November 3, 2025 at 9:46 PM
These days there are like 15-20 instances of Chrome+Node electron solutions running around on my OSX system for all my “apps”.

Progress! 🧐
October 30, 2025 at 2:30 AM
How much 💰 was lost by SaaS vendors 😢 today who had performance clauses for
their customers they missed
by a mile because of AWS outages?

Before you “well actually…” & say Amazon will be on the hook take a beat. Sure maybe some had SLAs but I’d wager those had redundancy too most folks didn’t.
October 21, 2025 at 1:13 AM
Reposted by Thomas A. Powell
Dead on. The better you are as a frontend engineer, the more you dish off to the underlying systems rather than keeping logic in your own code, which means that a big pile of JS landing on the main thread is generally a sign the org producing it is relatively unskilled.
Once you realize the web is really just a giant game / graphics engine that gives you a DSL for content, and a DSL for styling, and a scripting language for anything extra...it all kinda starts to make sense.

the "scripting" language is the escape hatch for the cases the DSLs can't handle.
October 20, 2025 at 5:34 AM
Does it feel like we should use LLMs like it is ending days of VCs propping up food delivery?
October 15, 2025 at 4:34 AM
I try to reframe thinking of an LLM tool as a form of info/code dynamite. 🧨 it is powerful and destructive and also quite useful in certain situations. Even in those moments though one must use and handle it with extreme caution ⚠️ as it can just as easily 💥 things as it make 🪄 ✨ things
October 12, 2025 at 4:50 PM
Today’s tech community seems to love to say open, decentralized, and protocol focused and really do a closed, centralized, and platform thing when control and money gets involved.

This isn’t just about this place. There are other examples.
October 6, 2025 at 8:22 PM
Reposted by Thomas A. Powell
It seems @jono.id is my Anger Translator:

jonoalderson.com/impolighthouse/

/via @tunetheweb.com
September 30, 2025 at 2:07 AM
Reposted by Thomas A. Powell
Glad Science collected this data (though the results are entirely unsurprising). GenAI cannot accurately summarize scientific papers, sacrificing accuracy for simplicity.

And shame on publishers who are pushing genAI summaries on readers. Great way to accelerate an epistemic apocalypse.
September 21, 2025 at 2:37 PM
Reposted by Thomas A. Powell
"In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limits."

www.computerworld.com/article/4059...
OpenAI admits AI hallucinations are mathematically inevitable, not just engineering flaws
In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limi...
www.computerworld.com
September 21, 2025 at 12:50 PM
Reposted by Thomas A. Powell
“I don't think anyone wants a Balkanized Internet, but I fear that's exactly where we're going.” – Steven J. Vaughan-Nichols

In their endless hunger for content, AI web crawlers are destroying websites. But the cure may ruin the web….

Hat tip: Dave Martin
AI crawlers destroying websites in hunger for content
Opinion: But the cure may ruin the web....
www.theregister.com
September 3, 2025 at 6:01 PM
Going to be teaching at USD this term as well as UCSD. Looking forward to their small class sizes. I’ll be really busy since I still have my industry job as well but important to try new things when you get a chance.
August 31, 2025 at 2:58 AM
When you see trending “Meta Ethics” what do you think? I’m sure nothing good.
August 30, 2025 at 2:49 AM
I say in response to those who think an LLM using written language alone will become sentient … “so how do you explain my dog and its’ intelligence? You are think language means intelligence!?”

LLMs leverage patterns to great value, but it’s as a path to AGI like a horse is a path to an airplane.
August 29, 2025 at 5:03 AM
So given this Subway Sandwich assault resulting in a felony charge does this then mean if you throw a McDouble at an agent they might throw you in Alligator Alcatraz? What about tossing a McFlurry is that straight to El Salvador? Just want to know what constitutes fast food assault of what degree?
August 14, 2025 at 5:10 AM
What Google Chrome is up for sale? I gotta get in on this. I’ve always wanted to own a browser.

Ok so my best and final offer is 100 Trillion Dollars. I’ll throw in some priceless Ape jpegs as a sweetener.

Ha ha I now have totally destroyed your crap low ball offer Perplexity!
August 13, 2025 at 3:07 AM
I propose we create a new test in the spirit of the Turing Test called the Altman-Musk Test. It measures human’s general gullibility and unquestionable faith in technology. You fail it if you believe the proclamations of these guys without question.
August 12, 2025 at 3:45 AM
Reposted by Thomas A. Powell
But my HTML websites from back then still work beautifully today, whereas my pals’ masterpieces can only be experienced on old computers running obsolete OSes, browsers, and plug-ins. So, there’s that. Better to have bread for 100 years than cake for 5 minutes. 2/2
August 11, 2025 at 3:44 PM
LLM companies are creating a loot box based knowledge economy and tokens are the in-app currency of the game obscuring the actual costs and influencing perception of value.

Sadly this post seems more reality than shitpost as of late.

#ai #llms #gpt5
August 9, 2025 at 6:40 PM
If #ChatGPT5 is supposedly on the way to AGI I can confidently say we humans are gonna absolutely wreck the Terminators it send at us.
August 8, 2025 at 4:59 AM
Reposted by Thomas A. Powell
Every front-end shop should have this printed out in huge type somewhere in a high-traffic common area:

muan.co/posts/javascript
July 16, 2025 at 7:43 AM