evrydayhustling

I'm a huge fan of both, but you're not the first person I've heard who had problems with the writing style / dialogue / characterization of Fire. In that respect, I'm not sure Deepness will feel different. Vinge has a pretty consistent style and cultural stance, which IMO is an interesting balance of Heinlein-style techno-libertarianism and genuine "humanism" (which extends to dogs, plants, and other thinking creatures).

Structurally, though, it's pretty different. The characters have much more opportunity to develop, and it's purely hard sci-fi where Fire used its setting to veer into fantasy ideas. So my biased advice remains: give it a shot :)

Metric time is about factors too, but it prioritizes the factors that simplify the comparison across (literally) many orders of magnitude. This is way less helpful for everyday life because we usually only reason within 2-3 orders of magnitude, but we deal with tons of harmonic subcycles that the 24h clock makes easy.

For a fun, a more justified usage of metric time, check out Vernor Vinge's "A Deepness in the Sky" (also IMO one of the best sci-fi novels of all time). A spacefaring humanity that stretches their life across journeys and projects that span centuries, and which has to artificially produce their own daily cycles, gets a lot more value out of metric time.

Those advances -- musket to Gatling gun, calculator to CPU -- are great examples of what our laws and customs must adapt to continuously.

In those cases, we could draw on analogy on things that were already possible for organizations - a machine gun is a one man army, a cpu does the work of a room of accountants. We have never had a Gatling gun for cogent paragraphs before, but we have contact centers full of scammers and propagandists.

A productive approach to controlling AI could start with considering our controls for these organizations (and how to make them work), and then address the concentration afforded by AI. Exceptionalism about what AI can do is a distraction.

> most of the problems of regulating AIs apply to regulating corporations, which can be thought of as AIs with slow internal data transfer

Or to regulating humans, who are like AIs with squishy bodies but still manage to build and manipulate dangerous stuff.

It is bizarre to me how doomist theses act like our society has never had to absorb autonomous unpredictable actors, yet that is exactly what society is made up of.

I agree that this is extremely valuable, but it's worth flagging that it's harder to reason about the impacts of class imbalance for generative models than e.g. classifiers. For example, should we think about genre imbalance per novel, per token, or on some more complex basis? Which genres are most relevant to a target distribution of chatbot queries?

This isn't to suggest that organizations shouldn't invest in actively understanding their training data, but that post-hoc bias analysis is going to be a critical component of evaluation for the foreseeable future.

I think "useful but late" is a tiny minority of the research that might come across your feed. The more common category is "advertising formatted as research".

There are many traditional money managers that would like to allocate to quant strategies because they are ostensibly uncorrelated with their existing investments. But the exact fact that these investors are traditional makes it difficult for them to access and evaluate meaningful quant opportunities. As a result there is a huge volume of aspirational research designed to attract attention and thus funding from this group, but which hasn't had any meaningful peer or market vetting.

So the take here is:

- I like how it feels to own a hammer, so everything should be like that. (I guess people shouldn't be able to rent hammers, or anything else?)

- You can't prevent the government from using what you build, so you might as well set up no barriers to anybody using it.

If you don't see any difference between limited control and no control, I don't think I'll convince you. But I think most of the ways we engage with the world involve degrees of control, and that there's value in picking where to exercise yours.

Asking builders to ONLY use maximally permissive licenses is equivalent to telling people never to release anything they're not ok with being used in every possible way. On a practical level this would massively chill research, as most builders and engineers I know give significant consideration to the impact of their work. On a personal level, it's objectifying: "Give me the code and don't make me consider your intentions in creating it."

You can't collaborate and live in a world free of others' value judgements, which are implicit in how they spend their time and what research / code / weights they choose to share. "Ethical" licenses at least make those value judgements explicit, and allow communities of builders with compatible values to organize and share.

For HN purposes, should remove the clickbait "oh my God". Headline speaks for itself!

I was particularly puzzled by the 2x2. High productivity is a no brainer, but why should we necessarily seek high displacement as its own goal? Low displacement would increase employment and GDP, is that not flourishing?

And what's with using the research scientists who authored Transformers as examples of surplus middle management? Google generated 1.4M per employee in 2022, second only to Apple. Armchair engineers will say they could do as well with a handful of employees, but if this is "low productivity" then sign me up.

Love this, tried it for ya. Check out the chat example [1].

For small changes, I'd rather read the diff. But occasionally, like item (10), it displays an understanding of the content and the summary is pretty useful. With guided prompts I bet you could make it much more useful, and even answer specific questions that align with folks' use cases for monitoring.

[1] https://chat.openai.com/share/c6ec81aa-de8c-49e3-b132-239136...

Whenever I see these gut biome results, I can't help thinking that there are so many common causes that can explain both biome and diagnosis/behavior: region, ethnography, socioeconomic class, etc. Has anyone gone deep enough into the studies to know if/how that is controlled for?

I like the analogy but I think there is an important difference: the issue settled in a boxing or MMA match starts and ends in the ring, but the issues covered in debate are social policies that depend on external truths and impact participants.

TFA gets into this in its last section about debate's impact on politics. Is it great to train a class of leaders based on competitive technique and effectiveness rather than external reality or support?

My first reaction reading this article was that this is indeed disturbing, but so was the format of debate I was exposed to in (turn of millennia) high school. The format basically embodies an idea that every opinion is equally valid, and what matters is how strong their advocates are.

Model UN, a close cousin, emphasized multipolar perspectives but in a context of research, creativity and cooperation.

I think the ethos of Debate is one of the drivers of today's polarization, long before Kritiks arrived.

True, but he followed the cardinal rule of successful traders everywhere: gamble with other peoples' money.

The life insurance front sounds like a lot more work than either the algorithm or the ticket buying logistics! I wonder how much of the payouts ended up with lawyers.

> More hypothetically, I would also expect that a world in which people spend a lot of time with screens strapped to their head, consuming an infinite stream of entertainment provided by generative AI, is not going to produce higher GDP.

Yeah, I think this dovetails with the idea that IT may be satisfying preference allocations without increasing overall production. Watching 10 movies a month on a streaming service adds much less to the GDP than going to the cinema 10x, but if the selection is better it might satisfy you more. Economists sometimes attempt to measure this with "utility adjustments" which recognize increasing quality in the same goods, but it's very hard for those adjustments to account for the hidden preferences of the consumers as opposed to objective qualities of a good or service.

Information goods like social media and streaming, financial services like pay-me-later, and conveniences like next-day delivery are all examples of activities that might suit preferences without showing up in GDP. They also may enable distraction, waste, reclusiveness and impulsiveness in ways we'd like to avoid as a society. At the same time they might also help some people feel more included and less lonely or trapped by circumstance.

The missing productivity paradox is something that should interest everyone in tech! Like any macroeconomic observation, there is plenty of uncertainty in both the measurements (GDP ain't perfect) and the context (2 financial crises and a pandemic). But even pro-status-quo institutions like Brookings agree that Something Is Up. Their 2021 review on the topic is both a decent summary and a good source of further refs [1].

My favorite explanation is that many new technologies end up redistributing wealth rather than creating it, which certainly tracks with both subjective and quantified growth in inequality on the same time period. However, a slightly more optimistic take is that tech is aligning production better with people's preferences, so that the same productivity enables people to live more distinct lifestyles that suit them.

[1] https://www.brookings.edu/articles/how-to-solve-the-puzzle-o...

There's a lot of preamble, but this part is pretty clear:

>The core of the case and our position is that Google abuses its control over the ad server monopoly to make it increasingly difficult for rival exchanges to run competitive auctions. Further, Google’s exchange rigs its own auctions so Google’s advertisers can buy ad space at bargain prices. That means less investment in online content and fewer ad slots for publishers to sell and advertisers to buy. Google always wins because it takes a growing share of that shrinking pie.

I'm unclear on which pie is shrinking. The sources I could find confirm that digital advertising revenues continue to grow and grow [1], while newspaper digital ad revenues have stayed constant (and print advertising has dropped steadily) [2]. Meanwhile digital subscription revenue -- some of which is presumably driven by search discoverability -- is growing nicely, including that of Gannett [3].

So the question seems more about "how much of the growing digital pie should newspapers receive" as opposed to "Google is building its revenues while destroying a market".

[1] https://www.statista.com/statistics/183816/us-online-adverti... [2] https://www.axios.com/2022/06/21/digital-newspaper-ad-revenu... [3] https://www.subscriptioninsider.com/type-of-subscription-bus... [3] https://www.subscriptioninsider.com/type-of-subscription-bus...

Some things I'd love to see:

- a difference metric between this coverage and the actual map (and country breakdown of same)

- a weighted version ([1]) with weights chosen to optimize the above distance metric.

The latter is in itself is a fun math problem -- the gradient on such an objective clearly exists (and the numerical gradient is fairly cheap). But can you prove anything about its structure? Is there a closed form that could be derived from the map information?

[1] https://en.wikipedia.org/wiki/Weighted_Voronoi_diagram

> Anyone upvoting this comment should take a long look at the names on this letter and realize that many are not conflicted.

The concern is that the most informed names, and those spearheading the publicity around these letters, are the most conflicted.

Also, you can't scan bio lines for the affiliations that impact this kind of statement. I'm not disputing that there are honest reasons for concern, but besides job titles there are sponsorships, friendships, self publicity, and a hundred other reasons for smart, "politically sophisticated" people to look the other way on the fact that this statement will be used as a lobbying tool.

Almost everyone, certainly including myself, can agree that there should be active dialog about AI dangers. The dialog is happening! But by failing to make specifics or suggestions (in order to widen the tentpole and avoid the embarrassment of the last letter), they have produced an artifact of generalized fear, which can and will be used by opportunists of all stripes.

Signatories should consider that they are empowering SOMEBODY, but most will have little say in who that is.

Since you are here, want to say this is one of the most useful posts I've seen about pragmatic development on top of LLMs!

And agreed re: development effort - compared to other hype cycles of AI, it's important for folks to understand that the results they see are coming at a fraction of the experimental budget.

The main thing LLMs can do is make products accessible/useful to a wider range of users - either by parsing their intents or by translating outputs to their needs.

This might result in a sort of transformation that engineers and power users aren't geared to appreciate. You might look at a natural language log query and say, "that would actually slow me down". But if it makes Honeycomb suddenly useful to stakeholders who couldn't before, it could lead to use cases not on the radar right now.

One approach: try to prove it doesn't have a solution. Would solving the problem entail things that obviously aren't true? Or does it entail things that probably aren't true, forcing you to dig farther about situations where they might be? Attempting to prove impossibility inverts the problem, and lets you act as critic on an argument that you want to be proved false.

This is a common approach to mathematical problems, but it's helped me break down more practical situations too.

The Vicuna headline was def an overreach, although the main text admits pretty readily that their performance test (asking ChatGPT to evaluate quality) is not rigorous [1]. I'm sure that has set a lot of people pontificating about AGI with tiny models, but I can't imagine anyone who has worked directly with fine tuning having that impression.

The comments I see here are not about that. They are about small models succeeding at specific tasks, which is affirmed by this paper. Most applications of LLMs are not general purpose chat bots, so this is not bad news for most of the distill/fine tune community.

[1] https://lmsys.org/blog/2023-03-30-vicuna/

That essay works in a context of specific datasets and tasks, which are referenced in the surrounding sentences and paragraphs. They are saying that for a particular "emergent" capability you might reach with a giant LLM, you might get there more efficiently with distillation / LoRa.

My comment is about generality, which is the remaining advantage of giant models.

This is a very weird type of paper. They take a specific approach, then make arguments about a broad class of approaches that are under constant development. The finding that distilled LLMs must be more specialized than the giant LLMs that train them is unsurprising; nobody at this point expects a 13B parameter model to succeed with the same accuracy at the broad range of tasks supported by what may be a 1T parameter model.

Then substitute "printing press" for "nukes" and we have mapped out the spectrum. Now where does AI lay? Today? Speculatively?

Bengio is a certified and maximally informed genius, so my thoughts obviously deserve far more salt than his. But...

The thrust of this argument is that centralized authority -- perhaps uniquely centralized, as Bengio proposes an agreement between Russia, China and the USA as an example -- is the best safeguard against rogue AI. He goes as far as calling out open source development of AI as a contributing threat:

> [if] the open-source community continues to play a leading role in the software development of LLMs, then it is likely that any hacker will have the ability to design their own pre-prompt (general instructions in natural language) on top of open-source pre-trained models...

The concern is logical, but the solution is self-contradictory (as many intellectually honest explorations turn out): a section later, we are appropriately exploring nefarious corporations as examples of mis-aligned intelligences -- and how larger corporations may be even more adept at circumventing the rules.

Whatever their philosophical foundations, our governments are also corporations (and so will be any new trans-national organization). They are fragmented into smaller groups with their own instrumental goals, and subject to the same risks of corruption that afflict private corporations, and the same risks of runaway optimization that may afflict AI. Appealing to them as a higher authority is at best a case of "the devil you know".

Doomism always has a non-falsifiable counterargument: AI is new and might become arbitrarily more powerful than the most powerful organizations we've ever created ("super"-intelligence). For me this fails on several counts: we are still exploring the limits of power (and of corruption) that large organizations can achieve, progress in AI remains observable as long as it is public, and our primary experience with AI has been colonizing a frontier of capabilities rather than a sequence of new systems dominating at every task.

To be clear, I do NOT personally conclude that we should avoid all regulation and oversight. But I think that constructing regulation around doomist theories serves disproportionately towards regulatory capture, and acts as a distraction from immediate issues of algorithmic inequality, disruption of political speech, and intentional weaponization of AI.

Our most crucial long-term defense against most of these risks is a deep detente, where the tools to understand and develop AI remain broadly available, so that no one group feels it has an unbeatable advantage in the powers afforded by AI. This is what the open-source community is currently providing. Locking down public research in AI in order to prevent superintelligence would have massive unintended consequences.

This title is clickbait. A direct title would be: "iOS photo permissions expose audio to applications".

This. Should be a browser plugin for maximum deployment impact.