One must think Sisyphus happy, and Cassandra on Prozac.
Rambles on:
#data #climate #genomics #politics #japanese #biking #electronicMusic #coding #bioinformatics #machineLearning #deOmniReScibili #✊🍉
Can be occasionally spotted in or near:
#tokyo #japan #london #uk #france
(posts self-destruct after a few weeks)
| Pronouns | he/him/yo |
| 🖋️ | https://cassandra-on-prozac.com |
For additional amusement factor: Galbraith wrote that in January 1987, warning that the world economy did not look all that sturdy, despite money-people's assurances that things would go up forever.
Turns out he was right, and it all came crashing down 10 months later.
"Nothing so gives the illusion of intelligence as personal association with large sums of money"
JK Galbraith wrote this in 1987, but this might as well be carved on the epitaph of our soon-to-implode world economy.
https://www.theatlantic.com/past/docs/issues/87jan/parallel.htm
Found out that my electric adjustable desk now only works *down* (up button doesn't do anything.
Had I realised it was truly broken and unable to go back up, I may have avoided cranking it all the way down while fiddling with it… 😑
How's *your* Tuesday morning going?
People are joking that #starmer barely outlasted the string of equally useless people who had the job before him, and that he will leave no legacy whatsoever.
But I think that's really selling short his many accomplishments: enabling transphobia, prosecuting genocide protesters, sucking up to Trump, and generally being the most pathetic enabler of rightwing policies ever elected on a leftwing mandate.
I guess what I'm trying to say is: it is blindingly obvious to anyone with even a cursory background in proper ML, that most LLMs have amazing language-related capabilities… and laughably poor accuracy on pretty much any measurable task.
Talks of "curing cancer [or climate change] with LLMs" are about as serious as the use of psychics or ouija boards…
One thing I can tell you from my years of reviewing bioinformatics papers: the vast majority of people (and a scary number of researchers) have an extremely poor grasp of model validation principles (statistical testing etc). If I had a token for every machine learning paper I read that thoroughly failed to account for their overfitting bias, I'd be running my own AI company…
The fact that the word "overfitting" has all but disappeared from modern AI is just spooky.
Ladies and gentlemen: the likely future Prime Minister of Great Britain… 😂
Do I know anyone who dabbles in secure decentralised/p2p protocols (MLS, hypercore… hell even activitypub)…
Someone who would potentially be willing to entertain a few stupid questions from a noob…
RE: https://mamot.fr/@pluralistic/116766875505739917
As someone who literally used to do “AI cancer research”, I feel endless pity for my former colleagues, who must not only contend with vanishing funding, but also with their entire field’s blend of advanced maths and systems biology, being reduced to spicy autocorrect technology, on a daily basis.
