I've read the Bible. Especially as you get towards the end, the books don't seem to have an especially lofty view of the criminal justice system.
I will agree with you that a criminal justice system built "on pragmatism" would certainly conflict with the tenants of many world religions. I recently read Reforming Criminal Justice: A Christian Proposal, which outlines why pure pragmatism needs to be tempered by a respect for, and indeed love of, every person accused or convicted of a crime.
Not GP, but I've definitely seen cutting edge LLMs make language mistakes. The most head scratching one I've seen in the past few weeks is when Gemini Pro decided to use <em> and </em> tags to emphasize something that was not code.
What's wild is that OpenAI's earlier models were trained to guess the next word in a sentence. I wonder if GPT-2 would get "though" correct more often than the latest AI-assisted writing tools like Grammerly.
There are some areas where it seems like LLMs (or even SLMs) should be way more capable. For example, when I touch a word on my Kindle, I'd think Amazon would know how to pick the most relevant definition. Yet it just grabs the most common definition. For example, consider the proper definition of "toilet" in this passage: "He passed ten hours out of the twenty-four in Saville Row, either in sleeping or making his toilet."
One of my friends is a public health lobbyist and she is used to having to explain to stakeholders that THIS formatting mark in the PDF means they're adding text but THAT formatting mark means they're deleting text. It's not immediately obvious and every state has its own way of presenting information. I'd argue that DC does it best, but I haven't looked at every legislature.
Another advantage is you can easily include deprecated models in your comparisons. I maintain our internal LLM rankings at work. Since the prompts have remained the same, I can do things like compare the latest Gemini Pro to the original Bard.
I'd be really interested in evaluating the evaluations of different models. At work, I maintain our internal LLM benchmarks for content generation. We've always used human raters from MTurk, and the Elo rankings generally match what you'd expect. I'm looking at our options for having LLMs do the evaluating.
In your case, it would be neat to have a bunch of different models (and maybe MTurk) pick the winners of each head-to-head matchup and then compare how stable the Elo scores are between evaluators.
I have a tangential question: how useful do you find SVM in production? I've generally found it to be too slow compared to other algorithms to be worth it.
I've experienced the same thing. I work with ad agencies, and it's common for my client to then turn around and present the same information to their clients, so they'll ask me to put every last word on the slide. It hurts my soul.
When I walk into a client's office for a meeting, I like to be able to plug into a projector and talk through my slides (which I export to PDF for simplicity). I don't want to have to ask for a guest Wifi password or anything else technical. As a result, I usually give presentations offline, unless the projector needs me to connect over the internet. Sure I could tether my laptop to my phone, but why add a dependency?
You nailed it. I imagine this is why OpenAI is looking to develop hardware. Right now, I have tabs open for ChatGPT, DeepSeek, and Gemini. I have zero loyalty to any of them. But if I owned a piece of hardware, I immediately am locked into that ecosystem.
I respect that. I graduated during the Great Recession. Ended up working for extremely low wages out of college and ultimately went to grad school after getting nowhere in trying to find a better position.
I really identify with your first paragraph. I enjoyed writing for the school paper and programing in high school, but I didn't see much of a future in either, for different reasons (death of print and offshoring). I graduated around the time that 538 launched a new appreciation for data-driven journalism, so maybe I could have been in a good position.
I ended up running the data science division at a digital publisher, so I guess I backed into that space after all.
Glad I'm not the only one who stayed away from Reddit after the initial boycott. I used to be a regular contributor with a bunch of front page posts. Being away made me m realize how much time, effort, and stress I put into building magic internet points. Now I spend that time doing things for myself instead.
I had a great boss who really liked that kind of question. I disagreed with him. I would rather have someone who knows how to find the official answer online and verify the quality of the source.
Oof, reminds me of the data pipelines I maintain that pull data from Salesforce. I've optimized the heck out of all of our internal database queries and they're blazing fast. Then my pipeline has to wait patiently for the SOQL response.
But but my GCP-PDE exam said every project needs massive and expensive infrastructure to optimize every tiny detail, minimize latency, and deliver five nines of availability! If your cloud bill isn't five figures every month, are you even a dev?
/s
Well said! At work, I deal with these kinds of issues, and they get messy. I've spent hours this week discussing and debating table indexes and caching parameters for a system that's been running for fifteen years but we think we can improve. There's a diminishing return to putting tons of your time into every little thing, especially when a project is not your livelihood. It sounds like you've taken a thoughtful approach to your system architecture. If it works for you, it works!
I have a tangential question related to one part of the article. I worked in a .NET shop where a surprising amount of code was VB.net (the rest was C#). Was this a common design pattern for .NET code written around 2010? The article makes it sound like nobody really used VB.net, and certainly not when there were C# devs around.
I completely agree. It baffles me that there aren't more entry level EVs in the United States.
Two years ago, I decided I wanted an EV. I didn't want to spend a ton. There weren't many good options. I was disappointed by the Model 3. Ionic was promising. Ended up buying a Chevy Bolt EUV. I absolutely love it. The torque you get from even a modestly powerful electric vehicle is incredible, and one pedal driving is a real boon in city driving. Plus it's fun to drive past the gas station and just plug into the charger that Chevy paid to install at my house.
I don't want most of my vendors sending me messages on Slack or Teams. If there's an issue that requires attention, please send me a thoughtful email with the relevant information.
I'm surprised more people don't use images for this kind of thing. You could put an image with a green checkmark next to every live job posting, and change it to a red X once the job is no longer available. I worked for a publisher of email newsletters and we ran a bunch of tests into the caching behavior of email clients to see how viable this kind of thing would be. Turns out it's pretty easy to implement and works just about everywhere.
I was on the opposite side. When AMP4email came out, I worked at a publisher of email newsletters. I pushed our CTO to implement it. He was less than enthusiastic. We ultimately decided it wasn't worth the effort and now that looks like a smart idea.
Agreed. The Wall Street Journal suggests setting the y axes so your data fits in 2/3 of the available space (unless you have a specific reason for a different setting, such as putting zero as the minimum for revenue).
I have the same experience in the web UI. Asking for that famous Obama chili recipe gets a refusal. But when I use the API, I can dial back the safety settings to the point where things work much more smoothly.
I agree with your specific point. I think copyright should last for 50 years. That would protect long running franchises like Harry Potter or the Dresden Files but still allow things like Spiderman and Aretha Franklin to be freely reimagined.
However, most of the examples in the article should still be protected. Lara Croft first appeared 29 years ago. Any copyright system should still be protecting IP from 1996.