chris's Avatar

chris

@chrisecramer.bsky.social

Sr Director of AI and ML / 30+ years with ML and computer vision. Beekeeper, home brewer, and general fermentation fan. LLM skeptic

238 Followers  |  638 Following  |  274 Posts  |  Joined: 18.09.2023  |  2.4197

Latest posts by chrisecramer.bsky.social on Bluesky

Maybe I've lost the point, but this is embarrassingly wrong. It's spitting out a 10x10 grid with 9 numbers between 1-10 (no 9 on the top row and no 10 on the first column) and the math is obviously wrong. What is this supposed to prove?

19.11.2025 11:30 β€” πŸ‘ 4    πŸ” 1    πŸ’¬ 2    πŸ“Œ 0

I'm pretty sure you can drag the new window back into the old window and the new window goes away and becoemes a tab again. It's definitely undoable. Now having way to turn off the feature for those who don't want it makes sense :-)

14.11.2025 17:14 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

So I guess from a technical standpoint, I would be cautious. And that doesn't even get into the jobs aspect

07.10.2025 17:52 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I've heard of other similar issues in translation software using genAI. IIRC, there was a recent story about K-pop demon hunters using ChatGPT to help with songs, but it turned out to be an artifact of the translation software

07.10.2025 17:52 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

For example, the medical transcription tool whisper (from OpenAI) has been shown to invent entire conversations, in part bc of weird context attention, and in part bc once it gets one thing wrong, then the whole transcription can go off the rails.

07.10.2025 17:52 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

The other form of AI translation is based on generative AI, think ChatGPT. These are predictive models that try to predict the next word based on what's come before and the current input. The problem with them is that they are less grounded in the original

07.10.2025 17:52 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Google translate is based on a sequence to sequence model that tries to map lang A to lang B. It is fairly accurate and grounded in the meaning of the original text. There are also seq2seq models for spoken to written text.

07.10.2025 17:52 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

It sounds like the goal is have spoken language translated into another spoken language? Or into another written language?

From a technical standpoint, there are two approaches that are broadly labeled as AI.

07.10.2025 17:52 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Agreed. And several of the examples provided (boiler plate code, website templates, etc) are probably better accomplished with an actual code template.

29.09.2025 20:19 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

It's like glitter...it's been hundreds of years and we're STILL finding Vikings everywhere πŸ˜†

23.09.2025 07:53 β€” πŸ‘ 104    πŸ” 23    πŸ’¬ 4    πŸ“Œ 2

And this is why you're the brains of the family - or at least the memory 😁

15.09.2025 23:08 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

You should! They're great

15.09.2025 01:22 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Near there - this was just up the trail from Rainbow Falls in Gorges State Park. It was July 4th week and we were avoiding DuPont 😁

15.09.2025 01:18 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Horizontal photo of a waterfall. The falls aren't very high, but are pretty broad. The camera speed is low enough that all of the splashing looks super smooth.

Horizontal photo of a waterfall. The falls aren't very high, but are pretty broad. The camera speed is low enough that all of the splashing looks super smooth.

Same trip to the mountains of NC, different waterfall. I can't remember the name of this one, but it was a bit of a scramble above another falls... whose name I also can't remember πŸ˜•

#photography #waterfall

14.09.2025 23:19 β€” πŸ‘ 104    πŸ” 9    πŸ’¬ 4    πŸ“Œ 0
Vertical picture of a waterfall surrounded in greenery. You can see the reflection of the falls in the water below

Vertical picture of a waterfall surrounded in greenery. You can see the reflection of the falls in the water below

Another picture of the same waterfall, but this time horizontal. The photo is cropped more closely to the falls which is surrounded by greenery. 

The water is very smooth as the picture was taken with (iirc) a 6 stop ND filter and had a 4 second exposure time

Another picture of the same waterfall, but this time horizontal. The photo is cropped more closely to the falls which is surrounded by greenery. The water is very smooth as the picture was taken with (iirc) a 6 stop ND filter and had a 4 second exposure time

Two shots of the same waterfall (Schoolhouse Falls) in the mountains of NC. I'm still not sure which I prefer.

#photography #waterfall

13.09.2025 21:43 β€” πŸ‘ 8    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Showing this to the LLM fanboys in my company and asking if this is their god

13.09.2025 15:59 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

As I understand it, yes. It could let the cultivar root and negate the benefits of the rootstock. But it's very possible there's something I don't know :-)

13.09.2025 00:55 β€” πŸ‘ 6    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I mean, to be fair, I would love to spend a few hours with Hedy Lamarr, discussing spread spectrum communication systems. I'm not sure that constitutes a date, but...

11.09.2025 23:35 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

"how can you get more accurate answers" from an LLM? that's like asking "how can you get more love from a prostitute." that's just not the service provided.

β€” Thomas' words of wisdom

07.09.2025 22:05 β€” πŸ‘ 391    πŸ” 65    πŸ’¬ 6    πŸ“Œ 3

Nice explanation!

02.09.2025 00:41 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Yeah - a trillion different possibilities is amazing... and now I want miso soup πŸ˜€

02.09.2025 00:02 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Which is a lot of misos

01.09.2025 23:58 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

So, my result:

G1: 10
G2: 171
G3: 13
G4: 102091
G5: 56
G6: 8

Total: 1,016,728,352,640

01.09.2025 23:57 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

No, I'm still wrong. One more time.

N choose K is N! / (N-K!) / K!

So N=18, K=1, gives 18 pssobilities

N=18, K=2 gives 18! / 16! /2! Or 18 x 17 / 2.

So that group has (18 + 18x17/2) total options

01.09.2025 23:49 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Group 3 has 1 + 12 (12 choose 0 is 1 and 12 choose 1 is 12)

01.09.2025 23:44 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

So group 1 has 10 possibilities. Group 2 has 18 + 17. (I was wrong above. Not possibilities to the power, but possibilities choose min...max)

01.09.2025 23:43 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Got it - so, your categories are independent and can be multiplied together. For each category, you have options raised to the min power + options to the min+1 + options raised to ... all the way to max

01.09.2025 23:41 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

Can you give a quick example of what you mean by minimum and maximum number of inclusions?

For category, I assume you mean that there are 6 types of thing that always go into the food, so maybe proteins, spices, carbs, etc?

01.09.2025 23:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

My sense of it is that the most readily available water for cooling would be from the municipal water supply. But I don't have the reference either

31.08.2025 18:54 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

WRT clean/filtered - as I understand it, most of the water use is from evaporative cooling and I would assume that anything that isn't clean/filtered is going to be a cleaning problem later when the water evaporates

31.08.2025 17:07 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

@chrisecramer is following 20 prominent accounts