Benjamin Smith

Computer scientist.
Supercomputer pseudo-scientist.
Former U.S. Navy cryptologic technician.

[Benjamin Smith @HashCons]

I have added new features to the NFL @Grok @Bot that provide Kelly Criterion values based on current prediction market values.

The bot also provides a custom "Weather factor" adjustment based on the current forecast on gameday.

Reports include a summary of injuries.

https://t.co/7BSrIzZn60

[Benjamin Smith @HashCons]

@clemycali @SanDiegoKnight Close enough! I'm sure it happened at different rates everywhere. It might have happened at my undergraduate also, but they just failed those people. I know our first intro class had about 50 students, and only around 12 graduated with us.

[Benjamin Smith @HashCons]

I think the schools just got addicted to the money. I went from being a student in a very difficult program to being a TA at a school where everyone who pays tuition gets to pass.

Some of us saw it, and some played along. I had students who were so illiterate, some couldn't even write comments in their code. Not typos, like actually failing to spell small words.

That was almost 20 years ago, and it's scary what they might be in charge of now.

[Benjamin Smith @HashCons]

An entire movie made from AI clips a few seconds long?

Hard to watch.

[Benjamin Smith @HashCons]

@DewsNewz We had to deal with garbage quality images and video for so long, all it takes is a little bit of blur and compression to fake people out.

[Benjamin Smith @HashCons]

@BrettAMacDonald So, a billion more Americans, has he really thought about that?

Assuming we have all those kids, they're still 100% useless for decades. At existing costs, that's trillions of dollars a month more just for public school.

[Benjamin Smith @HashCons]

I used to make stupid stuff like this all the time. In a world before AI. Now I can't imagine spending hours drawing all that.
https://t.co/CAg6OMIGpv

[Benjamin Smith @HashCons]

Integer factorization is really hard. So hard that recently the first RSA number since 2020 was factored. These numbers have been public for decades, and used to have huge financial rewards if anyone solved one. No "Quantum" there. Just regular nerds with far too much compute.

[Benjamin Smith @HashCons]

Find it a little strange that all those "Most experts" always seem to predict things just far enough in the future that nobody questions them.

Despite my experience, I'm not an expert, but I know enough to look for signs that something is off. Look at Shor's algorithm, a real technique for attacking real problems. Notice anything missing?

[Benjamin Smith @HashCons]

They're a little blurry because I had to take screenshots. Apparently X doesn't like SVG images.

[Benjamin Smith @HashCons]

Have you ever seen Petri nets displayed with so much valuable information? Probably not, @Grok just created it for me.

Beautiful!

[Benjamin Smith @HashCons]

There's far more capability available than people using phrases like "Vibe coding" are suggesting.

Admittedly, this project was already the most advanced solution in the world for this particular problem, and I don't think Grok or any other model could have generated it from scratch based on a simple prompt.

With that in mind, I emphatically recommend people start using these SI tools for the most advanced problems in their field. This didn't take a huge "Token budget" I am just using what I get with my basic $30/mo. combined subscription, with room to spare.

Making progress on problems where decades, and millions of dollars, have been spent is fun!

[Benjamin Smith @HashCons]

How about a project update?

With the @Grok usage set to refill tomorrow, I've put a large part of my existing code through a significant rewrite. Grok spread some of my code over multiple lines, and it's easier to read so I don't mind a higher count.

Next step is multi-threaded.

[Benjamin Smith @HashCons]

@witcheer A classic tech demo. Perfect mix of "I want one" and "In 6 months that's going to look so old."

[Benjamin Smith @HashCons]

@S2DioAlpha @grok It's hard to say. Like this function, mine's certainly shorter, and possibly faster. Mine runs through the loop twice no matter what, while grok's should just be once.

I'll have to test it on real problems to see if there's a difference worth spending more effort.

[Benjamin Smith @HashCons]

@frankiemacd WINNING!

[Benjamin Smith @HashCons]

@BillSchulz A real life werewolf in California, playing basketball, and he's the only one that thinks there's money to be made.

[Benjamin Smith @HashCons]

@Oilfield_Rando With a name like General Bacon, his true destiny was being a mascot for a fast food chain.

[Benjamin Smith @HashCons]

@Grok created the original cartoon image, just from text based on my recent X posts at the time, and I've always been impressed.

This time I asked to make the images real, trying the same two prompts.

My local model made me overweight and gave me stubble and sideburns. Grok gave me golf balls to eat once.

Never mess with an artist's vision!

[Benjamin Smith @HashCons]

I tried adding a hat to my profile picture using @Grok, and with my local model. Same prompt.

Of course, the text... but I think the local model got the cartoon style of the hat a little closer.