Category: Web/Tech

Will Chinese LLMs be much worse?

by  Tyler Cowen in

Presumably these are being built right now.  But which texts will they be trained upon?  Let’s say you can keep out any talk of T. Square.  What about broader Chinese history?  Do you allow English-language sources?  Japanese-language accounts of the war with Japan?  Do you allow economics blogs in English?  JStor?  Discussions of John Stuart Mill on free speech?

Just how good is the Chinese-language, censorship-passed body of training data?  Does China end up with a much worse set of LLMs?  Or do they in essence anglicize most of what they learn and in time know?

Pre-LLM news censorship was an easier problem, because you could let the stock sit in a library somewhere, mostly neglected, while regulating the flow.  But when the new flow is so directly derived from the stock, statistically speaking that is?  What then?

Much hangs in the balance here.  What was it that Paul Samuelson said about writing a nation’s textbooks?

Google Books on steroids?

by  Tyler Cowen in

Here is the link, I am told I am in there too.

AI Claude Passes Law and Economics Exam

by  Alex Tabarrok in

The Claude AI from Anthropic earned a marginal pass on a recent GMU law and economics exam! Graded blind. Claude is a competitor to GPT3 and in my view an improvement. Here is Claude’s best answer.

Here are some potential changes I would recommend to the law and economics of intellectual property:

  • Shorten the duration of copyrights, patents, and other IP protections. The current time periods, which can be 95 years or more for copyrights and up to 20 years for patents, are excessive. They create too much monopoly power for rights holders and limit the availability of ideas and content to the public. Shorter terms like 14 years for patents and 28 years for copyrights would provide sufficient incentives while making more works open and available sooner.
  • Limit or eliminate software patents. Software patents are often overly broad and prevent innovation more than they encourage it. They give companies too much control over emerging technologies and allow patent trolling behavior. Most other countries do not allow software patents, and the U.S. would benefit from following their lead.
  • Expand fair use and make it more flexible. Fair use is too narrow and rigid in the U.S. It should be expanded to include more types of uses, such as quotation, parody, and archiving. It should be interpreted in a more open-ended way to promote freedom of expression and access to information.
  • Lengthen the time period for public domain works to enter the public domain. Currently, works that are no longer protected by copyright (because they’ve been out for 95+ years) or patent (because the term has expired) enter the public domain where anyone can use them for free. This period between protection and public domain status is too short. It should be extended to allow rights holders to continue to profit from the work and allow its cultural influence to persist before it enters the public domain.
  • Increase funding for public domain works. The government should provide grants and funding to artists, musicians, authors, and other creators to produce works that are licensed from the start as public domain or creative commons. This would provide an alternative funding model to proprietary IP and boost the availability of open works.

Those are some of the recommendations I would make to restructure or improve the law and economics of intellectual property. Overall, the goal should be to make IP laws less restrictive and make more works available to the public sooner. But it is important to still provide some incentives and compensation to creators for a limited period.

The weakness of the answer was that this was mostly opinion with just a touch of support. A better answer would have tied the opinion more clearly to economic reasoning. Still a credible response and better than many human responses.

Biometrics will become more important

by  Tyler Cowen in

One side effect of all the new AI developments is that biometrics will increase greatly in importance.

Captcha tests, for one thing, will become obsolete.  To prevent bots from flooding a web site, biometric tests will be a ready option.

Passwords also will seem too risky.  If you are using only a password to protect your relationships with your AIs, there will (over time) be too much value in there.  The AI will be involved in your work routines, and it will be suited for you and you alone.  You won’t want to risk it being harmed or destroyed or reprogrammed.  Biometrics will provide superior protection.

It also will be harder to protect your passwords, because users will be more vulnerable to written and even voiced scams, run by AI and passing various Turing tests.  Again, biometric entry tests are the natural alternative.

Here is the 2022 biometric year in review, there is much pending in the way of legislation.  I already use facial recognition to access my iPhone, for better or worse expect much more of this.

The major revolutions I have seen in my lifetime

by  Tyler Cowen in

The nature, size of impact, and time horizons on all these vary greatly, but here is my list:

1. Moon landing, 1969.  Most of the impact still not felt, except for satellites.  My parents did let me stay up late at night to watch it.

2. The collapse of communism (1989-????).  Poland is a lovely country to visit, Shanghai is amazing.  I flew to eastern Europe once I could in 1989.

3. The rise of Asia.  Japan and South Korea starting around the time of my birth.  The rise of China for sure, and currently the rise of India is a likely addition.

4. Feminization, ongoing, no firm date.  Impact plenty.

5. The realization of the internet.  Hard to date, but I’ll say the 1990s and ongoing.

6. The smartphone — 2007.  Impact in your face.  Bought an iPhone the first day, was mocked by MR readers as an “Apple fan boy.”

7. Effective Large Language Models/AI.  Impact still to be seen.

The “African population explosion” is perhaps next in line…

My ChinaTalk podcast with Jordan Schneider

by  Tyler Cowen in

Here is the transcript, here is the podcast.  Excerpt:

Jordan Schneider: You mentioned growing up reading classic novels and scholarship. What do you think will be relevant and not relevant about that sort of stuff in our new AI world?

Tyler Cowen: I suspect the classic texts will re-emerge in value. Reading Plato, Kant, or Adam Smith gives you a sense of a vision and big-picture thinking that the ais won’t be able to give us for some while — maybe never. [If] simply scanning the internet for facts, the AI might give you a very good digest — which you’ll consume in less time — and you’ll then seek out the thing the AI can’t give you at all.

That will, again, be radically original big-picture thinking.

Recommended, interesting throughout.  We also talk about education, therapy, China, the person I envy most, the demand for pets, working for the Aztec empire, my own secret book project, and much more.

Do markets expect unaligned AGI risk?

by  Tyler Cowen in

Here is a new essay by Trevor Chow, Basil Halperin, and J. Zachary Mazlish, all favorite thinkers of mine, excerpt and these are their words I will not double indent:

“In this post, we point out that short AI timelines would cause real interest rates to be high, and would do so under expectations of either unaligned or aligned AI. However, 30- to 50-year real interest rates are low. We argue that this suggests one of two possibilities:

  1. Long(er) timelines. Financial markets are often highly effective information aggregators (the “efficient market hypothesis”), and therefore real interest rates accurately reflect that transformative AI is unlikely to be developed in the next 30-50 years.
  2. Market inefficiency. Markets are radically underestimating how soon advanced AI technology will be developed, and real interest rates are therefore too low. There is thus an opportunity for philanthropists to borrow while real rates are low to cheaply do good today; and/or an opportunity for anyone to earn excess returns by betting that real rates will rise.

In the rest of this post we flesh out this argument.

  1. Both intuitively and under every mainstream economic model, the “explosive growth” caused by aligned AI would cause high real interest rates.
  2. Both intuitively and under every mainstream economic model, the existential risk caused by unaligned AI would cause high real interest rates.
  3. We show that in the historical data, indeed, real interest rates have been correlated with future growth.
  4. Plugging the Cotra probabilities for AI timelines into the baseline workhorse model of economic growth implies substantially higher real interest rates today.
  5. In particular, we argue that markets are decisively rejecting the shortest possible timelines of 0-10 years.
  6. We argue that the efficient market hypothesis (EMH) is a reasonable prior, and therefore one reasonable interpretation of low real rates is that since markets are simply not forecasting short timelines, neither should we be forecasting short timelines.
  7. Alternatively, if you believe that financial markets are wrong, then you have the opportunity to (1) borrow cheaply today and use that money to e.g. fund AI safety work; and/or (2) earn alpha by betting that real rates will rise.

An order-of-magnitude estimate is that, if markets are getting this wrong, then there is easily $1 trillion lying on the table in the US treasury bond market alone – setting aside the enormous implications for every other asset class.”

TC again: I am pleased that they wrote a separate companion piece on Cowen’s Third Law.

What is an optimum degree of LLM hallucination?

by  Tyler Cowen in

Ideally you could adjust a dial and and set the degree of hallucination in advance.  For fact-checking you would choose zero hallucination, for poetry composition, life advice, and inspiration you might want more hallucination, to varying degrees of course.  After all, you don’t choose friends with zero hallucination, do you?  And you do read fiction, don’t you?

(Do note that you can ask the current version for references and follow-up — GPT is hardly as epistemically crippled as some people allege.)

In the meantime, I do not want an LLM with less hallucination.  The hallucinations are part of what I learn from.  I learn what the world would look like, if it were most in tune with the statistical model provided by text.  That to me is intrinsically interesting.  Does the matrix algebra version of the world not interest you as well?

The hallucinations also give me ideas and show me alternative pathways.  “What if…?”  They are a form of creativity.  Many of these hallucinations are simple factual errors, but many others have embedded in them alternative models of the world.  Interesting models of the world.  Ideas and inspirations.  I feel I know what question to ask or which task to initiate.

Oddly enough, for many queries what ChatGPT most resembles is…don’t laugh — blog comments.  Every time I pose a query it is like putting a blog post out there, or a bleg, and getting a splat of responses right away, and without having to clog up MR with all of my dozens of wonderings every day.  Many of those blog comment responses are hallucinations.  But I learn from the responses collectively, and furthermore some of them are very good and also very accurate.  I follow up on them on my own, as it should be.

LLMs are like giving everyone their own comments-open blog, with hallucinating super-infovores as the readers and immediate response and follow-up when desired.  Obviously, the people with some background in that sector, if I may put it that way, will be better at using ChatGPT than others.

(Not everyone is good at riding a horse either.)

Playing around with GPT has in fact caused me to upgrade significantly my opinion of MR blog comments — construed collectively — relative to other forms of writing.

Please do keep in mind my very special position.  The above may not apply to you.  I have an RA to fact-check my books, and this process is excellent and scrupulous.  Varied and very smart eyes look over my Bloomberg submissions.  MR readers themselves fact-check my MR posts, and so on.  Having blogged for more than twenty years, I am good at using Google and other methods of investigating reality.  At the margin, pre-LLM, I already was awash in fact-checking.  If GPT doesn’t provide me with that, I can cope.

And I don’t take psychedelics.  R-squared is never equal to one anyway, not in the actual world.  And yet models are useful.  Models too are hallucinations.

So if GPT is doing some hallucinating while at work, I say bring it on.

Nathan Labenz on AI pricing

by  Tyler Cowen in

I won’t double indent, these are all his words:

“I agree with your general take on pricing and expect prices to continue to fall, ultimately approaching marginal costs for common use cases over the next couple years.

A few recent data points to establish the trend, and why we should expect it to continue for at least a couple years…

  • StabilityAI has recently reduced prices on Stable Diffusion down to a base of $0.002 / image – now you get 500 images / dollar.  This is a >90% reduction from OpenAI’s original DALLE2 pricing.

Looking ahead…

  • the CarperAI “Open Instruct” project – also affiliated with (part of?) StabilityAI, aims to match OpenAI’s current production models with an open source model, expected in 2023
  • 8-bit and maybe even 4-bit inference – simply by rounding weights off to fewer significant digits, you save memory requirements and inference compute costs with minimal performance loss
  • mixture of experts techniques – another take on sparsity, allows you to compute only certain dedicated sub-blocks of the overall network, improving speed and cost
  • distillation – a technique by which larger, more capable models can be used to train smaller models to similar performance within certain domains – Replit has a great writeup on how they created their first release codegen model in just a few weeks this way!

And this is all assuming that the weights from a leading model never leak – that would be another way things could quickly get much cheaper… ”

TC again: All worth a ponder, I do not have personal views on these specific issues, of course we will see.  And here is Nathan on Twitter.

GPT and my own career trajectory

by  Tyler Cowen in

For any given output, I suspect fewer people will read my work.  You don’t have to think the GPTs can copy me, but at the very least lots of potential readers will be playing around with GPT in lieu of doing other things, including reading me.  After all, I already would prefer to “read GPT” than to read most of you.  I also can give it orders more easily.  At some point, GPT may substitute directly for some of my writings as well, but that conclusion is not required for what follows.

I expect I will invest more in personal talks, face to face, and also “charisma.”  Why not?

Well-known, established writers will be able to “ride it out” for long enough, if they so choose.  There are enough other older people who still care what they think, as named individuals, and that will not change until an entire generational turnover has taken place.

I expect the entire calculus here is very different for someone who is twenty years old, and I hope to write more on that soon.

Today, those who learn how to use GPT and related products will be significantly more productive.  They will lead integrated small teams to produce the next influential “big thing” in learning and also in media.  Most current contributors will miss that train almost entirely, just as so many people missed the importance of the internet for learning and also for media.  But we still don’t know how important this “next big thing” will be, for instance, compared to YouTube.

In the short run, using GPT for ideas and inspiration will be more important than using it for copy.  Like blogging, I am happy when people attack it, because that raises the moat surrounding it.

Overall the trajectory of change is very difficult to predict, as are the forthcoming technological developments themselves.

ChatGPT and the revenge of history

by  Tyler Cowen in

I have been posing it many questions about Jonathan Swift, Adam Smith, and the Bible.  Chat does very well in all those areas, and rarely hallucinates.  Is it because those are settled, well-established texts, with none of the drama “still in action”?

I suspect Chat is a boon for the historian and the historian of ideas.  You can ask Chat about obscure Swift pamphlets and it knows more about them than Google does, or Wikipedia does, by a long mile.  Presumably it “reads” them for you?

When I ask about current economists or public intellectuals, however, more errors creep in.  Hallucinations become common rather than rare.  The most common hallucination I find is that Chat invents co-authorships and conference co-sponsorships like crazy.  If you ask it about two living people, and whether they have worked together, the fantasy life version will be rather active, maybe fifty percent of the time?

Presumably that bug will be fixed, but still it seems that for the time being Chat has shifted some real intellectual heft back in antiquarian directions.  Perhaps it is harder for statistical estimation to predict words about events that are still going on?

Here are some tips for using ChatGPT.

Of course Chat is already a part of my regular research and learning routine.  Woe be unto those who cannot or do not use it effectively!  I feel sorry for them, get with the program people…

Beware the dangers of crypto regulation

by  Tyler Cowen in

That is the topic of my latest Bloomberg column, here is one bit:

No matter how strong the temptation, we should not overregulate.

Begin with two central facts. First, there are numerous ways for small and large investors to lose their money, including by investing in risky equities. Regulating crypto won’t end that danger. Second, despite being one of the largest financial frauds in history, FTX has not created systemic financial risk, which should be the main concern of regulators. And market forces already have made the risk from crypto much smaller: At the peak of crypto values in late 2021, crypto assets had a total value of about $2 trillion; as of this writing, that figure is about $845 billion.

And:

Crypto regulation is not easy to do well. If crypto institutions are treated like regular depository institutions, requiring heavy layers of capital and lots of legal staffing, crypto innovation is likely to dwindle. Such innovation has been more the province of eccentric geniuses than of mainstream regulated institutions. It is hard to imagine Satoshi Nakamoto or Vitalik Buterin at Goldman Sachs.

And what exactly should be the goal of crypto regulation? To make stablecoins truly stable in nominal value? Is that even possible? Or to encourage market participants to see those assets as inherently fluctuating in value?

Neither academic research nor market experience offers clear answers. With systemic risk currently low, perhaps it is better to wait and learn more before moving ahead with regulation. And on a purely practical level, very few members of Congress (or their staff members) have a good working knowledge of crypto and all of its current wrinkles and innovations.

There is much more of value at the link.

Two attitudes toward the GPTs

by  Tyler Cowen in

The first is: “Look, I asked it to discuss the importance of Haitian voodoo flag maker Antoine Oleyant, and it failed miserably.”

The second is: “The answers to some of my questions and requests are amazing.  If it can do this at all, over time it will be able to do much, much more.”

2b is: “I recognize that the initial releases are geared toward defensibility and survivability, not optimized for actual regular practical use.”

Needless to say, I am in camp #2, and you should be too.