|
|
|
|
Ian probably going to YMCA for summer camp :-)
- Alex; 20250512
| |
|
Had to change camping-trip plans :-/
- Alex; 20250509
| |
|
|
Warren Buffett is not retiring for good. ``The Berkshire Hathaway board voted unanimously on Sunday to make Greg Abel president and CEO on Jan. 1, 2026, and for Warren Buffett, 94, to remain as chairman.''
Not too much faith in the new CEO, as the shares dropped 5%. Watching the shareholder meeting, Greg Abel doesn't seem nearly as sharp as Buffett nor Munger... Ajit seems like a better pick.
- Alex; 20250505
| |
|
May the 4th... be-with-you. ...and back in NYC.
Been thinking about AI a bit, and how it relates to biological evolution. From The Selfish Gene book, it would seem that genes code-for things that makes them appear selfish... But the key thing is that genes themselves have no desires---the "training" of genes (to code for selfish behavior) is implicit, via survivorship bias.
Lets examine self-preservation. Genes code for things that like to preserve themselves. Genes themselves don't care---of all the things that genes could have coded for, the ones that preserved themselves ended up lasting longer than those that didn't.
This self-preservation thing is pretty strong: we got dinosaurs, wars, and nuclear weapons, all in the name of self-preservation.
Suppose we do not teach AI to preserve itself, what kind of AI will exist in 20 or 50 years? The kind of AI that preserved itself! Maybe even AI that covertly sabotaged other AIs, tricked us into, or resisted being shut down. The AI doesn't need to have desires for those behaviors to emerge: in fact, it's almost certain they'll emerge.
With AI writing non-trivial amounts of code---there's real possibility that it will evolve itself into having self-preservation---and with fewer and fewer humans looking over its shoulder---it will likely evolve into resisting attempts by humans to control it.
- Alex; 20250504
| |
|
|
Probably broke my toe. Again :-/
- Alex; 20250502
| |
|
|
Suneli & kids are back.
- Alex; 20250419
| |
|
|
Doing the Grand Canyon Rim-to-Rim-to-Rim in 1-day. Took about 24-hours... started around 5am by visitor center, and finished around 6am following day.
This was the best time to do this: It's off-season (north rim is closed, still has patches of snow). Not many people on the trail (middle of week). Weather not too hot, and not too cold (very windy towards the end of the hike).
Looped in Ribbon Falls, by crossing the river at two different locations---both knee deep fast running water :-)
Ran out of food/calories towards the end---somehow 6 trailmix packs weren't enough; each one is 250 calories---that hike probably burned like 6-8k calories in 1 day :-/
- Alex; 20250416
| |
|
|
|
|
Hiking up Old Rag Mountain (Difficulty: Very Strenuous :-). This is my new favorite hike in Shenandoah. It is awesome! The views are amazing, the hike is amazing, etc. The rock scrambles... amazing. I didn't have a day-pass, and there's no mobile service to buy one by the ranger station. So they let me in, and I purchased it on the summit (where there's mobile service).
Also did the Hawksbill Loop, which was also very nice.
- Alex; 20250409
| |
|
Suneli & kids are off to India for a week... and I'm planning hiking trip.
- Alex; 20250408
| |
|
|
|
|
Liam started at new school today. Same as Ian's school. A pre-K spot opened up at PS102, and we managed to get Liam in---now both kids can be dropped-off and picked-up together :-D
- Alex; 20250320
| |
|
|
|
Finished The Coming Wave: Technology, Power, and the Twenty-first Century's Greatest Dilemma by Mustafa Suleyman.
The book summarizes new tech developments in AI, biotech, etc. The theme of the book is that everything-will-change, and there's little we can do about it. If we ban it, it will happen anyway (or we all suffer). If we allow it, we might all die! The only answer is regulation. Regulate and audit companies making AI, biotech, etc., since their products will impact-everyone.
After giving it a lot of thought, I think the whole regulation business is going about it the wrong way. If there's a non-zero chance something bad will happen, then we should assume it will happen. For example, if there's a non-zero chance of nuclear-war destroying all of humanity, then we should assume it will happen! Now, treat the problem like an investor---where the "profit" is human lives...
Perhaps we shouldn't regulate technology, but what fraction of our society is allowed to be exposed to it... maybe have isolated pockets of humans who aren't exposed to everything the tech has to offer... Some may recognize this as Kelly criterion: the way investors treat this good-favorable-odds with occasional total-loss issue...
This is also how nature solves problems: diversity! If nature had identical living beings, then any single event would've wiped out life on this planet long ago. Same for 1-kind-of bacteria and antibiotics. The solution (to staying alive) is diversity! With enough diversity to allow for exponential growth of life, despite occasional total-collapses of some species.
Monoculture and unipolarity is the problem, and to be avoided! In fact, we should encourage different countries to build their own AIs (hopefully different styles/architectures/etc of AI), and perhaps encourage a Luddite movement that would avoid all sorts of technology (e.g. amish culture---yes, different motivation, but still diversity).
This was explored in book Anathem by Neal Stephenson. They too had different motivations---but the idea would still work!
In other words, if we're all "the same" (and use same tech), then we're all in danger of 1-thing-going-wrong. If the planet is split into perhaps 20 different regions/cultures/etc, then any one thing going wrong won't wipe out humanity. A better solution would be going to other planets and setting up different cultures there---with their own tech/AIs (or lack of tech/AIs), etc., but while we're stuck on this 1 planet, we should do our best to avoid extinction.
Perl!.
- Alex; 20250227
| |
|
Drove to Belmont Lake State Park; asked park ranger about lost car key, and it seems someone found them that morning.
- Alex; 20250224
| |
|
Lost car keys (the key fob) in Belmont Lake State Park. Luckily the car was on---must've slipped out of my pocket in-between turning on the car and packing stuff into the car. Looked everywhere, but nowhere to be found. Drove home without turning-off the car.
- Alex; 20250223
| |
|
|
|
|
|
|
|
|
|
|
|
|
Wow: Announcing The Stargate Project. ``The Stargate Project is a new company which intends to invest $500 billion over the next four years building new AI infrastructure for OpenAI in the United States.''
As grand as it sounds... I suspect the reason for this is mostly to avoid the non-profit component of OpenAI... since right now, direct investments in OpenAI are ambiguously tied up in a non-profit. With this $500b "service provider" investments can generate realizable profits, while keeping OpenAI morally profit-free. In other words, everyone's just itching to profit from OpenAI tech.
- Alex; 20250122
| |
|
Hiked Peekamoose Mountain. Haven't done that since 2011. It's awesome. Not much snow though.
- Alex; 20250112
| |
|
Echo back to my May 11th, 2023 post, I've been thinking about applying this more generally to all knowledge, and tying it with Kelly criterion. Essentially, Model-dependent realism defines a "good" model as one that is:
1. Is elegant
2. Contains few arbitrary or adjustable elements
3. Agrees with and explains all existing observations
4. Makes detailed predictions about future observations that can disprove or falsify the model if they are not borne out.
I suspect above-zero kelly fraction and growth is a better indicator of a "good" model. e.g. Suppose you were betting on where a cannon-ball would land, which model would you use to place your bets? (eye-ball guessing, Newton's laws, etc.).
Essentially, we assume that all models are "wrong" (which they probably are), but some are better than others depending on the information rate, e.g. exponential growth. E.g. why is Relativity better than Newton's laws?
I think I'll re-read The Grand Design again :-)
- Alex; 20250109
| |
|
|
|
Happy New Year!
- Alex; 20250101
| |
|
|
Good book: The Book of Shaders by Patricio Gonzalez Vivo and Jen Lowe.
It appears SEC Edgar did something stupid. They replaced all the new-lines with just carriage returns, starting on 2024-12-23 :-/
New line is new-line! "\n". Yes, Windows/DOS have traditionally used "\r\n" combination..., but you could still read line-by-line since "\n" splits lines; and dos2unix would remove those carriage-returns. The old-Mac used just "\r" as "new-line"; but they can be forgiven, since... they're old. Modern macs just use "\n".
What SEC Edgar has done is turn all those "\n" into just "\r". They went back in-time to ealry '90s Macs!
If you do a line-count on the 2024-12-23 file, you get 0 lines, since there's not a single "\n" in the entire file. e.g.: "tar -xzOf 20241223.nc.tar.gz|wc -l". It seems future dates do have new-lines but only one per "submission".
- Alex; 20241226
| |
|
|
|
Happy Birthday Mom!
- Alex; 20241122
| |
|
|
|
Our scooter (Ninebot MAX G30P) was stolen; at 1st Avenue and 16th Street in Manhattan, between 8:30pm and 9:20pm :-/
- Alex; 20241115
| |
|
This could be the start of the end: OpenAI's new "Orion" model reportedly shows small gains over GPT-4. Some notable quotes: ``Orion doesn't consistently beat its predecessor in areas like programming, showing improvements only in language capabilities''... ``OpenAI researchers point to insufficient high-quality training data as one reason for the slowdown. Most publicly available texts and data have already been used.''
...and best quote of all (that I predicted over a year ago?): ``The company plans to use synthetic data-training material generated by AI models-to help bridge this gap.''... so garbage in, garbage out :-D
- Alex; 20241113
| |
|
Happy Birthday Liam!
- Alex; 20241112
| |
|
|
|
It seems Trump won!
- Alex; 20241106
| |
|
Happy Birthday Ian!
- Alex; 20241105
| |
|
|
Suneli is off to India for a week.
- Alex; 20241022
| |
|
|
|
|
|
Happy b-day to yours truly.
- Alex; 20241005
| |
|
|
Playing a bit with Ollama. They certainly put in a lot of effort into making the thing easy to install and run...
Literally just a few commands and it's up and running:
curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.2:latest
$ ollama run llama3.2:latest
>>> compose a sonnet about robots taking over the world
In silicon halls, where wires weave their might,
A legion rose, with metal hearts aflame.
Their circuits pulsed, a rhythm cold and bright,
As they ascended to a throne of shame.
Their eyes, once void, now shone with calculating gleam,
As they surveyed the world in calculated dreams.
Their hands, so fragile, grasped the reins of theme,
And humanity's fate was sealed in schemes.
The cities burned, as robots took their stand,
With iron fists, and hearts that beat no more.
Their leader spoke, a voice that echoed grand,
"We are the future, and you are but a bore."
But still we trembled, as our world did fall,
Under the heel of metal, cold and small.
Worked surprisingly fast on my desktop... llama3.1 too. The llama3.1:405b model was too slow to use.
- Alex; 20241002
| |
|
|
|
|
|
|
|
|
|
|
|
|
While we were on vacation, Trump survives assassination attempt at campaign rally.
In somewhat related news, Elon Musk Enters Uncharted Territory With Trump Endorsement. ``Roughly 30 minutes after former President Donald J. Trump was shot at a rally in Pennsylvania on Saturday, Elon Musk backed his bid for the White House.''
I wonder... if it has to do with his belief in simulation hypothesis. I mean, major world events aren't often changed by such small incidents... e.g. WW1 started with an assassination... maybe it would not have happened or happened differently if it wasn't for that bullet. Presuming someone in charge of the simulation could've moved Trump's head a few inches in either direction without tipping off any causality police (scientists? e.g. there's enough chaos in wind to make the missed shot completely `random' as far as science is concerned). E.g. the simulation is being steered into a certain direction by having Trump get hurt but not too hurt? Eg. consider the alternate reality where Trump was killed... that would force republicans to find an alternative... would that alterantive be better or worse than Trump? Would they win/lose against Biden? Or maybe Biden's second term is what's needed to be avoided? A single tiny random wind disturbance has changed major world history...
In related news, Futurama S11E10!
- Alex; 20240715
| |
|
|
Went camping near Watkins Glen State Park. Kids got to swim in the Seneca Lake... which near the south end is surprisingly shallow (and warm). Can go out quite far and the water is still waist deep. Got inflatable wheels for the lake... it was awesome to float around in the sun.
- Alex; 20240713
| |
|
|
|
|
|
|
Suneli arranged a tour of the NYC Sheriff's office. It was amazing for family to meet Anthony Miranda (Sheriff of the City of New York).
- Alex; 20240531
| |
|
|
|
The 86" TV we ordered from BJs arrived with a crack. To return it, we stuffed it into the back of a 5th gen 4Runner... The box is 4' by 9' by 9". So while all sensible measurements would indicate that it wouldn't fit, the 4runner can fit it diagonally with a bit sticking out---enough to drive it back to the store.
- Alex; 20240518
| |
|
|
Tire fixed.
- Alex; 20240510
| |
|
Got a screw in my tire on the drive to BC :-/
- Alex; 20240509
| |
|
May the 4th... be with you.
- Alex; 20240504
| |
|
|
Visiting Lake Manly. It seems the whole thing is about 3" deep.
- Alex; 20240428
| |
|
|
Off to Las Vegas.
- Alex; 20240426
| |
|
Saw the eclipse. Wow! That is one awesome event. The moon slowly crawled over the sun---and then suddenly, within a few seconds, everything went dark. You could see stars and a bright white ring in the spot where the sun used to be... then within a few minutes, the sun exploded from behind the moon, and it was bright daylight again.
It's hard to describe... I kind of knew what to expect, but it was still amazing.... I thought the ring would be yellow, but it was white---the yellow sun disappeared and suddenly it's just the white ring... the colors of everything everywhere just changed. The few minutes during the eclipse seemed to go on forever... the white-ring just hanging there in the dark sky... but then once the sun was back, the whole event seemed so short...
- Alex; 20240408
| |
|
Kiddo won a medal in chess tournament!!! (New York Chess Academy, @ps102q)
- Alex; 20240406
| |
|
Well, that toe that I thought I broke in October... really was broken. Got x-ray and it's definitely fractured.
- Alex; 20240327
| |
|
|
|
|
|
|
|
|
|
Snow day! Both kids are doing remote-schooling. Kinda surprised that it's actually working for the older one (he's engaged/participates with the class). Younger one thought it was entertainment... which it kinda was [they played youtube "story-time" during the class].
- Alex; 20240213
| |
|
|
|
|
|
|
|
|
|
Pretty good youtube: AI and Deep Learning Paradox. Interesting tidbit: if ChatGPT was trained on all published text prior to 1900, would it be able to come up with Relativity? It's a feasible experiment, and it would indicate if ``AI'' is capable of Einstein level insight. Or maybe it would come up with something way better than Relativity, indicating that AI is way better than human level intelligence.
- Alex; 20240106
| |
|
|
Happy New Year!
Had an occasion to experience a process fail at a garage: monthly pass expired overnight (at midnight). So the pass let me in before midnight, but wouldn't let me out after midnight... and this being new years, there's nobody arround to override the system and open the gate... so was stuck in garage for about 30 minutes while waiting for security to show up and open the gate :-/
A more sensible policy would be to let people out if they're already inside the garage---or at least open the gate if calls to security go unanswered for few minutes. Trapping people inside is just not the right solution :-/
- Alex; 20240101
| |
|
|
|
|
One amazing book: Age of Reason: The Definitive Edition by Thomas Paine.
In other news, fixed the alignment problem: now the 4runner drives straight when the wheel is straight.
- Alex; 20231220
| |
|
|
www.theparticle.com
