Makes me wonder idly, - is this conceptually akin in some sense to a "universal grammar," and if so - with a broad enough training set, is there a latent durable universal grammar that might be similarly recovered and injected to the benefit of all training, - does that grammar go beyond morphological/syntactical/grammatical features, into e.g. semantics and pragmatics

Even in the target-rich environment of late-stage kleptocratic surveillance capitalism, it is hard to pick out a less ethical firm which has done more moral and direct harm, to more people, with as durable a record of defying all attempts and internal and external at harm reduction.

Never use their products. Never partner. Never participate.

Black/red is a hard no for me. Something fitting with the existing aesthetic (let's call it "Scandanavian milled", subtle frosting, white text) or warmer ("natural", off-white, orange LEDs) etc etc.

One other question: waterproof? I assume that's not possible. Would be a decision maker for me personally probably.

When the "AI community" GTFO X and stays off.

Toxic site. Toxic ownership. Unbelievable bot activity. Indefensibly shitty politics constantly boosted.

Continued participation is a stain on every company and person who continues to use it.

There are alternatives. Don't like them? Make a better one.

Stop using that shithole.

but this is just lovely and inspirational work. A unique marriage of hacking in the celebratory sense, thoughtfulness, and meaningful contribution. Outstanding.

To sharpen that, I think he's (obviously) interested in maintaining his own brand as "thought leader" and this necessitates de rigeur defense of particular postures.

Sometimes this is easy because the facts warrant it; other times, a bit of rhetorical license is required to preserve nominal coherence and (at least, for the moment) hold certain lines.

This is one of the latter cases, and it's not subtle.

One of the celebrated properties of many intellectual advances or inventions in whatever domain is precisely that it appears obvious in hindsight. It is quite cynical to leverage consensus distrust of large AI players, warranted but also a popular social construction, to insinuate that these are not "real" advances or "real" hard problems, on the grounds they were in some sense cherry-picked.

Identifying the problems amenable to strategies on the table and intuitions (sic) about where bridges might be, is exactly the discerning work that is the core driver of almost all prior progress, but for celebrated accidents and flashes of insight. Anyone working in any challenging discipline knows that those are celebrated and told around campfires precisely because meaningful durable results arising like that is so uncommon.

These two articles make me think of nothing so much as my own durable reaction to the creeping goalposts of AI critics generally: that they often seem to me not unlike a water color cohort scoffing and jeering at the horse, because it got a D on its tensor calculus exam.

Marcus should be on guard against his own cynicism and take care that his assumptions do not prevent clear sight.

I have all praise for those taking this on and in my idiom would call it *the lord's work."

The image I reliably summon to mind is that compilation video showing the progress of Boston Dynamics bots. The curve between technically functional, to comically slow, to too slow for "real" work, on to, OMFG, may prove a (rough) curve.

It's work like this that moves things forward.

This line of critique is pernicious because it is both technically correct, as description, and profoundly misleading.

Saying that outputs are a product of inputs is not interesting and to the point it is not explanatory.

What is interesting, is how they do what they do. What is the "statistically likely* next token? To answer that you can do exactly one thing, run the LLM. That's because what they are doing is interesting and not reducible.

What is more interesting is that in order to do what they do, given the architectures we apply and the training strategies we use and the harnesses we situate them in, LLM are recapitulating in their deep layers strategies observed in the animal brain. This is still suggestive, interpretibility is nascent: but it is also more than a little interesting. In some respects, for cognitive scientists interested in the manner in which mind merges from computational substrates, it is profoundly interesting.

One can incorporate this, and, still be viciously critical of bother the success and failure of LLM in the applications we have put them to, and of how we (as individuals and as institutions such as corporations) are integrating them into our work.

There is a lot to criticize!

But criticism can be taken more seriously when it is not obscured by misunderstanding or misrepresentation (intentional, or not) of what LLM are and why they are not remotely "parrots" in the pejorative sense.

The technology, as technology, at the scale we are architecting it, is doing things we did not imagine would be witnessed in our lifetime, if ever. Dismissing that and denying it because of the career, industry, society, and civilization challenges that technology brings are existential, is bad argumentation or bad faith.

Both can be true at once.

Especially given its occurrence in the summary paragraph, that is a strong claim and I would like to know more.

Some googling turns up social media links expanding on this claim ("reflective and insulating") but not immediately anything authoratative.

It's part of his GUT. You will likely become committed to the well when you encounter ruliad: https://writings.stephenwolfram.com/2021/11/the-concept-of-t...

This depressing observation is regularly voiced (or thought) every time UX aesthetics come up... and I've yet to see so much as a sniffy defense.

My assumption this is because it's simply true; the only reason for churn is "fashion" requiring orbiting around a small space of viable options, strange-attractor style, and the vested interest an industry has in fueling the very modest pleasure that comes from tiny or bold steps from what is familiar to what is "fresh."

which is that capable models running in environments where they can be and regularly are abliterated,

and as a society and civilization, we have no good answer to what to do about the destabilizing threats this poses.

I don't have an answer.

And I'm not arguing in favor of some cynical solution which distills down to the state defending oligopolies.

But the threats are real and we have to reason about them soberly, widely, loudly, and quickly.

Most readers here know that defense is always at a disadvantage when it comes to security. That is true to an extent hard to fathom given the innumerable dimensions along which an antisocial actor can apply the force-multiplication of capable models to ill ends. It doesn't have to be FUD-adjacent concerns such as biological terrorism and cybersecurity, though these are real; the opportunities for mischief and misadventure are endless and the opportunities for poisoning the body politic or bringing down basic infrastructure many.

What then is to be done...?

I don't know; but I do know that the argument that almost all use of open models poses no threat is insufficient.

We have not reckoned as a culture with force-multiplication such as this technology brings.

We need to get ahead of the curve, and that will require making novel hard and unwelcome decisions in the short term.

The IMO inevitable outcome otherwise is to rue in leisure, after some global shock, assuming we have opportunity to.

I am 100% not on team "quietly/secretly update the text" eg in ebooks...

...I was more thinking of this as an interesting "art" project of sorts, adjacent maybe to an annotated edition where readers can opt in to a different kind of experience.

For me the most interesting bit wou;d be the problem-space of how to "translate" the cultural signification across time, which I think we would agree is literally (heh) impossible and a dubious and Quixotic goal.

I think it was in The Mind's I that the open question of whether in translating Tolstoy one could and should re-situate the novel in London for English readers, or, re-situate Dickens novels in Paris for the French edition, etc...

...all of this just forces one to confront literary theory in a productive way.

Because I enjoy such theory and also absurdism it does feel like it would be fun to try in particular to produce an automated PKD-esque black-box automaton which duly tries via "AI" mechanisms to produce candidate translations for us of exactly those aspects of something like Neuromancer which are most significantly coupled to the time place and framing of its writing,

the landscape of features that were most problematic (comical!) to translate would I think also be a neat heatmap reflective of our cultural evolution from/diversion from that time place and framing...

It's common for games to have an intentional designed aesthetic which uses cultural and social and graphical "languages" to create specific emotional impacts and effects. The lower the budget the more this is leaned into for expedient reasons but that in turn creates genre expectations! Low-poly and 8-bit are fine examples of what were once a product of technical limitations becoming their own target.

Subject matters also, and the overall aesthetic of the game; triple-A games often push towards what you're describing, that the most impactful and compelling may also be the most real...

...or at least, contain elements of it. Lots of games mess with color shading and saturation etc for the impact-reasons...

I do look forward to the era of more "photorealistic" rendering as you prefer, especially for natural landscapes like those I enjoy IRL. Some of which can have very "unnatural" feeling swings in tone and all sorts of things! I was just up in the mountains this week and saw a lot more variation in lighting and its impact on a consistent landscape (mountains across a lake) than I've ever seen in a game engine so far :)

I immediately liked the game and was inspired to try multiple days,

my single biggest frustration was that once a day is attempted, it's locked, and, you can't continue after a single failure.

The "single point of failure" having heavy consequences would likely make me bounce off and not return.

Might I suggest an alternate design to try:

Keep the incentive to speed, but reduce the punishment.

Proposal:

Score is points-per-word as a function of time to complete.

With a non-linear asymptomic point curve.

Bonus for very fast, long tail decreasing from 1.x points to 0.x, after 30s you get 0 points...

...but critically you are not loocked out of the game and can take as long as you like.

Bonus for caching this state so you can return later

There are variations, e.g. a total time budget for all words... perhaps you retain a sense of tier, with players able to select a "league" which gives higher reward for fastest responses but more quickly drops you to zero for the rest of the game...

You can still have a leader board (implicit or explicit) where points are fungible.

You could even then have a "no stress" league which gives a flat 0.25 points per success with no timer at all or something.

No idea about either, myself.