David Meyer retweeted
This poem is a lipogram using only the 13 letters shared between English and Ukwehuwehnéha (Oneida language). It’s part of a series examining language suppression and Oneida/Haudenosaunee survivance. Here’s an audio reading of it.
0
1
1
36
A lipogram by @cerena with a meaningful constraint:
For your Native American Heritage Month reading pleasure, I am *very excited* to have a poem featured online and in print as part of the @AmPoetryReview Nov/Dec 2026 issue. “even the word extinction / can’t be / written in Oneida”
aprweb.org/poems/even-the…
0
0
1
71
.@nsmithdt, Crystalyne Curley, Chris Deschene win Navajo Nation president, vice president race navajotimes.com/reznews/crysta…
“Curley becomes the first woman elected to the tribe’s highest political office.”
“Curley becomes the first woman elected to the tribe’s highest political office.”
0
0
2
96
Next lecture on modeling gene regulatory networks in my Intro to Math Modeling class @UCSDMathDept @UCSDPhySci @UCSanDiego: fields, rings, ideals, and the ring of polynomials over Fâ‚‚.
Lectured on the lac operon in my Intro to Math Modeling class @UCSDMathDept @UCSDPhySci @UCSanDiego before starting to build a model for gene regulatory networks. Needed substantial biology background!
1
0
3
267
David Meyer retweeted
I always appreciate an opportunity to emphasize that the burgeoning quantum industry in California and elsewhere is building on decades of curiosity-driven fundamental research, mostly done at universities.
caltech.edu/about/news/cal…
caltech.edu/about/news/cal…
3
3
28
5,339
David Meyer retweeted
There's been a lot of hype (and criticism) around the recent Weight-Space Programming results: rikiyat.github.io/blog/weight-sp…
I looked into the actual setup. Using one network to edit another’s weights is old (hypernetworks go back to the 90s). The more interesting part is the technique they used to get a frontier model to do it.
They ran the model through Codex with a local harness: a bubblewrap sandbox that gives it a restricted Python environment (no network/GPU, persistent /work folder, only load()/save() for the weight tensors). Forward passes are kept completely outside the sandbox as a separate, budgeted tool, the model has to write special request blocks, and each evaluation (games or positions) costs one of a limited number of passes.
The result is a working controller for the target task. With a tiny evaluation budget, it produced networks that are much stronger than random initialization and competitive with networks trained the traditional way. Doing the same with standard RL would normally require millions of environment steps plus the usual actor-critic machinery and hyperparameter search.
This looks especially useful for settings where interactions are expensive (e.g. physical robots).
I looked into the actual setup. Using one network to edit another’s weights is old (hypernetworks go back to the 90s). The more interesting part is the technique they used to get a frontier model to do it.
They ran the model through Codex with a local harness: a bubblewrap sandbox that gives it a restricted Python environment (no network/GPU, persistent /work folder, only load()/save() for the weight tensors). Forward passes are kept completely outside the sandbox as a separate, budgeted tool, the model has to write special request blocks, and each evaluation (games or positions) costs one of a limited number of passes.
The result is a working controller for the target task. With a tiny evaluation budget, it produced networks that are much stronger than random initialization and competitive with networks trained the traditional way. Doing the same with standard RL would normally require millions of environment steps plus the usual actor-critic machinery and hyperparameter search.
This looks especially useful for settings where interactions are expensive (e.g. physical robots).
meh, the more you dig into each experiment, the less magical it gets; it feels like a loose collection of cherry-picked demonstrations with a grand narrative built around them
Exp 1-2 are tiny (~20K params) with unusually easy to interpret architecture; Exp 1 is an MLP, where nearly all params are used in the first layer, so you can easily tell what kind of objects the network is trying identify
Exp 5 is about designing a game strategy (we already know that Astra is extremely good at this) and then converting it to a tiny (~20K params) MLP; entirely doable by a human, but of course would require a lot of effort
Exp 3 & 8 are closer to automated mechinterp using python and established techniques, rather than direct weight interpretation (at least according to chat)
to be clear, those are genuinely cool results, but I think people are reading too much into them
I don't think any of this should meaningfully change anyone's intuition about what models are capable of
0
2
8
407
.@MarshSongs, “Imagine who they could indict?!”
And the California 98 shirt is a nice touch!
And the California 98 shirt is a nice touch!
Two years after our "Bohemian Trumpsody" #Queen parody (on Trump's presidential election, with footage from the Capitol attack - see below), here's our take on Freddie Mercury's high-energy belter, twisted to the forthcoming #Midterms, which we're calling:
👉"Go, Stop Him Now!"
0
0
1
90
Kojima &al. @CurrentBiology, Locomotor kinematics and head stabilization in rolling snakes (w/video) cell.com/current-biolog…
This is so ridiculously cool I checked to see if it is April first!
This is so ridiculously cool I checked to see if it is April first!
0
0
0
84
Lectured on the lac operon in my Intro to Math Modeling class @UCSDMathDept @UCSDPhySci @UCSanDiego before starting to build a model for gene regulatory networks. Needed substantial biology background!
0
0
2
378
Han, “my lab can move beyond simulation and run our quantum AI algorithms for drug discovery on real quantum hardware, testing where quantum genuinely adds value over classical methods”
I’ll save you the effort: now, it does not.
I’ll save you the effort: now, it does not.
Heart disease is the leading cause of death worldwide.
Cambridge researcher Namshik Han is on a mission to eradicate it completely.
He has co-founded a company, CardiaTec Bio, which is using AI and the potential of quantum computing to help make that happen.
Discover more for the future of healthcare 👉 bit.ly/4ySAErs
2
0
5
839
.@TerellWright2, “Families with substantial assets, like a vacation home or a yacht, might not be eligible for full tuition assistance, even if their income falls below the college’s threshold.”
So advertised thresholds are false?
So advertised thresholds are false?
Williams, Swarthmore and Grinnell are all cheaper to attend than their states’ flagship schools. A growing group of elite liberal-arts colleges are giving massive financial-aid awards to families making $200k + a year. It’s becoming one of higher ed’s best bargains. My latest:
0
0
1
196
David Meyer retweeted
What does it take to build a new kind of university?
As OIST marks its 15th anniversary, the people who have shaped its journey gather to reflect on the challenges, breakthroughs, and aspirations that continue to define the institution. Full discussion: youtu.be/-r962LjWYSo
As OIST marks its 15th anniversary, the people who have shaped its journey gather to reflect on the challenges, breakthroughs, and aspirations that continue to define the institution. Full discussion: youtu.be/-r962LjWYSo
1
4
21
1,704
.@NenadJD & @MTatalovic, Serbia just elected the world’s brainiest parliament science.org/content/articl…
"Student List–Students Are Winning has 141 representatives in Serbia’s 250-member National Assembly. Dozens of them come from academic world & more than half have a PhD or MD."
"Student List–Students Are Winning has 141 representatives in Serbia’s 250-member National Assembly. Dozens of them come from academic world & more than half have a PhD or MD."
0
0
2
138
David Meyer retweeted
In lieu of flowers, memorial contributions may be made to:
Johns Hopkins Alzheimer’s Disease Research Center Fund  Â
alzresearch.org/giving/
Alzheimer’s Foundation of America Caregiver Resources  Â
alzfdn.org/donate/
Photo Credit: Tommy Garcia
Johns Hopkins Alzheimer’s Disease Research Center Fund  Â
alzresearch.org/giving/
Alzheimer’s Foundation of America Caregiver Resources  Â
alzfdn.org/donate/
Photo Credit: Tommy Garcia
6,676
79,729
345,437
11,871,735
Very sad to see that Capt. Jean-Luc Picard has died.
bbc.com/news/live/cq23…
bbc.com/news/live/cq23…
0
0
5
209
That’s a lot of words for being speechless!
If you read the report you’ll find that the difficulty level of the reading passages has *increased*, making this a more complicated story than “the SATV is now less rigorous”.
If you read the report you’ll find that the difficulty level of the reading passages has *increased*, making this a more complicated story than “the SATV is now less rigorous”.
I am speechless. Thomas Aquinas College just dropped a study on the rigor of the new SAT.
It compares the 1980/87 test with 2026. The findings should terrify every parent and teacher...
Vocabulary has collapsed.
Old SAT: 70.6% of questions tested vocab.
New SAT: 15%.
The old know a wide range of words through analogies, antonyms, and sentence completions. The new test abandoned that. Most vocab questions now just ask students to fill in one blank from context.
Reading passages also got radically shorter (average ~97 words vs 300–340).
The new test uses short expository passages and hands you the necessary background knowledge inside the text. The older tests leaned much more heavily on argumentative and persuasive writing, assumed far more historical, scientific, and cultural knowledge, and penalized students who lacked it.
This is a redefinition of what "college readiness" means.
If you believe that kids should develop rigorous vocabulary, be capable of sustained reading, and possess background knowledge, I have bad news for you. The SAT no longer measures these.
If you want a future where children can actually read difficult texts and construct complex arguments, the SAT must be replaced now.
This cannot go on.
3
0
6
432
