Larry O'Brien

They Sherlocked #Cleft: voice note transcription -> summary just a feature in #Apple Notes come September.

For traveling, I like to be able to order off the menu in the native language. I got sick of #Duolingo's gamification and tried #Pimsleur for French. That's what I'd use if I were spending significant time there: Pimsleur is hard but no nonsense. I'm now on #Babbel but it's even easier than Duolingo and doesn't feel like I'm advancing at all. I might go back to Duolingo and submit to the gamification. Thoughts?

My my, they really made some interesting decisions for the finale of #TheCurse, didn't they?

Started S2 of "Slow Horses" last night. Were it not that last year brought us the stellar "Bad Sisters," S1 would have easily topped my list for best TV new for me in 2023. As it is, I'm at least delighted that there are 3 Slow Horses seasons in the can and 8 or more novels. I know they've greenlit an S2 of Bad Sisters, but that will be a tricky act. #TV #CouchPotato

#Doom was released OTD 30 years ago. It, and Wolfenstein, were so mind-blowingly good from a technical perspective that Xander Antoniades wanted to just flat-out interview Carmack and Romero. _Software Development_ magazine didn't do "interviews," and "business numbers." I wrote an editorial calendar, clipped potential advertisers, commissioned some articles, and the first issue of Game Developer hit the streets in April 94. #ComputerMagazines #GDC #gamedev

You may be cool, but you'll never be "broke my back when my plane's engine exploded and got taken to the hospital by a passing cowboy," cool. #ChuckYeager

I think it was Cory Doctorow who came up with the metaphor of corporations as "slow #AI." The #AlignmentProblem can be seen with corporations: there's a gap between what you want the system to do ("optimize societal benefit") and how it pursues that goal ("maximize short-term profits"). At the media level the gap is between "be rewarded for entertaining people" and the pursuit of "maximize engagement." If aligning "slow AI" has led to big problems, what about when AI becomes "fast"?

I'm surprised neither #Copilot nor #ChatGPT4 can write #Quicksort in #REXX. Might be a good testbed for learning how to fine-tune, e.g., #CodeLlama. (Assuming CL doesn't know REXX either.) How much code for syntax to "take"?

I was thinking of doing a one shot #hillfolk/ #dramasystem evening with the NPC crowd the party has encountered (enslaved dwarves primed to revolt. I think it would be awesome if the players developed the community leaders their #DnD characters have just met). I like the more narrative style of role-play, and would like to introduce it to the table, but it doesn’t seem like hillfolk is very conducive to a single evening of #ttrpg. What do you think?

Is anyone working on incorporating #FuzzyLogic into #LLMs? There was a brief period in the early 90s where “#Neurofuzzy” aimed for relevance, but nets were still shallow and fuzzy logic fizzled (AI Winter? Used but no longer called AI? Not competitive with existing control system formalities?). My #ShowerThought is that FL *feels* like it’s differentiable (continuous truth/belief values). If true, might be suited for integration with statistical machine learning. #ML

How does #LLM attention solve the #Winograd scheme "The bowling ball did not fit in the suitcase because it was too [big/small]."? Is this a specific case of having memorized the example?

In your #ScheduleC #Taxes, where do you put software subscription costs? Just "Office Supplies" or, I dunno', "Utilities"?

Tentative conclusion: I’m not going to get the minimum number of photos I need to reidentify a #mantaray below about a half-dozen. Once I have that few k-shot I get good results but I’m just not getting a “general model of how to tell one manta from another.” #MetricLearning #ArcFace #SubcenterArcface #Reidentification #LowKShot

Why I put so much effort into that #BostonWhaler. Taken yesterday. #Humpback #Hawaii

Now, I'm no expert, but I think my 1971 #BostonWhaler may have a #termite problem. (Oh, and the white thing is a #gecko egg.) #Boat

Interesting article about D-Wave "#quantum" annealing computers. https://arstechnica.com/science/2023/01/companies-are-relying-on-quantum-annealers-for-useful-computations/

My niche in the aughts was "#scheduling" optimization (90% of the time it was bin-packing, technically). I started with #SimulatedAnnealing and #GeneticAlgorithms, but generally found approximation algorithms (Vazirani's book) to be most practical. PS It's a great niche. #SoftwareDevelopment

@Krita This is the wave of the near-future. I bet similar "insta-scraping" will be done by all commercial "knowledge worker" tools, e.g., Word, Visual Studio, Figma, etc. What was "instrumentation" justified to find bugs is grist for the deep-learning mill. #AIEthics #SoftwareDevelopment #Instrumentation #Writing #KnowledgeWorker #Creatives

Any examples of using SIFT (or SURF or ORB) in #OpenCV to retrieve similar images from a DB? All the code I see is just comparing two images’ features at a time. If you have 1,000 images and you want the closest, what do you do? #cv2 #image_search