thethimble

How is this a stretch? As a percentage of the demographic population, A is decreasing and B is increasing.

The fact that there’s fewer people in the current demographic cohort accounts for much of the “missing” ~2m students.

Human teachers make mistakes too - perhaps even at higher rates than GPT in some cases.

Instead of isolating a single factor of hallucinations I think you have to also consider the cost and experience quality and make a more holistic judgement on whether these models are good or bad.

Not so “universal” if it’s per-observer, right?

If you think about the speed of light as the speed of causality then it is happening “now” in the sense that its effect is perceptible to us now. This seems like a good definition of now.

Grammar becomes complicated when relativity dictates that there’s no universal ordering of events/timeline.

Sure but react isn’t a monetizable product.

Fair, but the question isn’t whether it is useful but whether it will replace traditional search

The immutability guarantees let you build some really powerful constructs. Check out the world of immutable/functional data structures: https://en.wikipedia.org/wiki/Purely_functional_data_structu...

The book Righteous Mind by Jonathan Haidt was really interesting and relevant to your point. Haidt claims that every moral framework has a set of sacred beliefs that are not unlike religious beliefs.

In other words every framework has a set of foundational axioms that have to be accepted on faith.

The response quality of BlenderBot is pretty bad compared to ChatGPT

Example question: What’s a covariance matrix?

BlenderBot: Principal component analysis is a popular technique for analyzing large datasets containing a high number of dimensions/features per observation, increasing the interpretability of data while preserving the maximum amount of information, and enabling the visualization of multidimensional data.

ChatGPT: A covariance matrix is a symmetric matrix that describes the covariance between multiple variables in a given dataset. The diagonal elements of the matrix represent the variance of each variable, and the off-diagonal elements represent the covariance between each pair of variables. The covariance matrix is used in multivariate statistics to understand the relationships between different variables and to perform statistical inference and hypothesis testing.

These “terrible people” seem to have catapulted the world into a generative AI world.

They genuinely believe they will build AGI and therefore becoming the world’s most valuable company is a natural consequence.

Whether this is possible/probable is a different story, but I think a capped profit structure makes logical sense for the company that is aiming to create AGI. Would you want this technology instead in the hands of a for profit company?

It seems like the fine tune dataset to go from GPT -> ChatGPT is pretty valuable, particularly because it is proprietary.

Still, I agree with your characterization that we should see many similar models over time. As an example, see Deepmind’s Sparrow: https://www.deepmind.com/blog/building-safer-dialogue-agents

I think what’s clear is that this is an unprecedented type of use. I’m really interested in seeing how the courts rule on this one as it has wide implications for the AI era.

What’s the advantage of learning this positional embedding vs hardcoding one that has the desired properties of location similarity?

Perhaps I’m misunderstanding but what happens if you’ve had a one-off production issue (job failed, etc) and you hadn’t dynamically logged the corresponding code? You can’t go back in time and enable logging for that failure right?

It seems not too crazy to have a morse code format to input moves too, right? Knight C3 is just a few squeezes.

It’s level 3 on pg’s hierarchy of disagreement - the whole concept of which I find interesting: http://www.paulgraham.com/disagree.html

> Do they consider gold medals as a sort of confirmation that their bloody regime is doing good ?

Of course. The Olympics have been a way to signal a country’s prowess for generations.

https://apnews.com/article/sports-virus-outbreak-afghanistan...

Exclusively licensing GPT-3 to Microsoft seems like a clear example of this.

https://www.technologyreview.com/2020/09/23/1008729/openai-i...

I think if you explicitly type in a leading zero with five digits it seems reasonable that it should infer it’s a zip code.

To add to your point I think the competitiveness will increase over time with further technology innovation. Cloud providers are constantly adding features and trying to catch up with each other. New entrants are adding novel competitive dynamics as well (e.g. Cloudflare or Deno Deploy).

The cloud market is so large ($ TAM; also growing rapidly) I think there will always be a tailwind of investment and innovation that prevents monopolistic stagnation.

Well it’s not clear that anyone could have predicted the timing of this downturn. Just a few months ago people were painting a very rosy picture for the future. In that context, metering investment can have a huge opportunity cost as well.

I’m not advocating for CB - just suggesting that it’s harder to predict the future than you are implying.

Downside is limited to your principal investment.

Another interesting angle is that everyone eventually will become old and therefore in this protected class. It's not like race, gender, etc. in that way.

DDIA by Martin Kleppmann is fantastic.

Great job of providing low level details contextualized in a high-level narrative. He’s also got these amazing fantasy maps of the fictional landscape of data architecture: https://martin.kleppmann.com/2017/03/15/map-distributed-data...

I think you’ve forgot the websites with walls of white-on-white keywords/links at the bottom, trying to game pagerank. Many of these models were marked improvements compared to the generations before.

It’s not a perfect analogy but it’s the same reason we have laws that require seatbelts. In theory this impinges on the liberty of adults, but in practice the it obviously saves lives.

When people commit harm to themselves out of ignorance, I think it’s reasonable for a government to step in.

Note that I think it’s reasonable to fundamentally disagree from a libertarian standpoint but I don’t see many reasonable people arguing against seatbelt laws.

To state it directly, the vast majority of the population doesn’t have the capability of assessing risk in speculative investments and is disproportionately susceptible to fraud. You give the example of “investing in startups” without also describing that most startups fail.

> you fail to note the creation of a market happens by destroying another

This isn’t true. If you go back to the origin of human civilization, economic “output” was limited to survival - food/shelter. Today survival represents an ever-smaller fraction of economic activity. Where did this additional activity come from if it was limited to replace preceding activity?

Similarly real gdp per capita has been increasing since the beginning of time. How can this be possible if the economy is zero sum?

You’re being downvoted because you're conflating two concepts that are related but not the same: the money supply and economic output (GDP).

Economic output is non-zero sum. Consumption/production scales up and down elastically.

The money supply is adjusted independently in order to drive “ideal” output growth while maintaining “ideal” inflation levels.

I’d be very surprised if this is emotional thrashing from someone who has been very long-term focused and first principles oriented in all of his other endeavors.