jacobsimon

… It somehow just dawned on me that firm is between “hard” and “soft”

I think you’re missing the broader point, which is that there is a lot to computer science outside of the purely mathematical formalism.

For example, distributed systems and networking are more like a physical science because they seek to make generalized learnings and theorems about real world systems.

The author’s last point around complexity theory also resonates because it demonstrates the value of designing experiments with real-world conditions like computing hardware speed and input sizes.

My unprofessional take: The SEC is concerned primarily with protecting investors. If anything, changing to a normal for-profit structure and removing the cap on returns would be viewed as more investor/market-friendly than their current structure, which is partly to blame for what unfolded last year.

This is the way

Was thinking about this recently. The author misses the importance of timing in the unity of perception. It would be rather easy to conduct an experiment and ask people, “did X and Y happen at the same time?” where X and Y are different stimuli. You could test over a variety of different senses and time differences to determine if people are integrating their experiences or not.

The other important element is attention and awareness. You can certainly be focused on one thing more than another—this can be a useful kind of disunity

One thing I noticed going through school is that math concepts are usually taught first before physics and other subjects —- precisely because the math is viewed as a prerequisite for the other material. But this always seemed entirely backward to me, because much of the math was invented for and motivated by people trying to solve actual problems in these other disciplines. I think we should teach people in the same order of operations, rather than treating math as an abstraction to be learned by itself.

I know it’s not a completely fair comparison, but to me this question is kind of missing the point. It’s like asking “Why take a cab if you know where you want to go?”

The paper linked here starts off by asking whether LLMs can “solve complex problems in ways that resemble human thinking.” Why would we try to answer that question without discussing how humans think?

For starters, the naming is much less confusing. But the behavior also appears to be enforced/validated at some layer (hopefully?), which function calling did not seem to be. I was experimenting with it a couple weeks ago and it would work like 75% of the time but would often give me invalid results for schemas with relatively simple nested objects.

Oh wow didn't know that existed. You should try cmd-p too! If you hold down Cmd and press P repeatedly, it acts like ctrl-tab, but you also have the option to search files at the top. I suppose you could remap it to ctrl-tab and then you'd get it for free.

This is cool - I haven’t really thought much about the spatial organization of my IDE before. Reflecting on it now, I realize I don’t even use the tabs in VS Code anymore because I have way too many files open. I just look at one file at a time, and when I need to switch I use Cmd-P to search or jump between recent files. Works well but requires me to keep the names of files/directories in my head all the time.

Thanks for following up - just saw this!

Sorry I know what you mean - I’m just having trouble finding any data source / website that shows useful comparisons going back that far - what website are you looking at?

These are the 10 year returns I’m seeing according to S&P:

S&P 500: 11.1%

S&P 500 Top 50: 13.2%

S&P 500 Top 10: 18.1%

The trend is pretty clear, at least in the last decade. My other point was that the historical data may not be as relevant because the index is much more top-heavy today, perhaps in part due to the popularity of index funds and part because the top companies are actually outperforming.

Here’s an article to that effect - the top 20 stocks in the S&P accounted for ~90% of the index’s gains last year: https://www.visualcapitalist.com/cp/top-20-stocks-sp-500-ret...

To your question about picking the extreme Top 1 stock - again I’m having trouble finding good data on that, but I’d be curious what the hypothetical outcome would have been over the last 20 years.

I don’t know for sure, but I have a hunch this strategy has done below-average for the past 10-20 years, because most of the above-average returns have been driven by growth stocks with low dividends. Stock buybacks have also become a really popular way of returning value to shareholders instead of dividends.

Do you have data to back that up?

The best approximation I’ve found is S&P has this Top 10 index[1]. Over the last 10 years it has performed 18% annually vs 11% for the overall S&P 500, but that’s obviously been a historic bull run in large cap growth stocks. I can’t find data going back to 2000 to see how that strategy would have played out, but curious if someone else finds it or crunches the numbers.

1. https://www.spglobal.com/spdji/en/indices/equity/sp-500-top-...

Yeah to clarify, I don’t necessarily recommend (or even follow) this strategy, I still mostly invest in passive index funds.

One of my big brain investing ideas is to pick the stocks at the top of the index instead of buying the whole index. If index funds continue to rise in popularity, the stocks that are at the top will benefit most from passive investment volume.

Plus, index funds follow a kind of Pareto principle where the top stocks contribute disproportionately to the total return anyway.

As I’ve gotten older though, one of my realizations is that the tax-free rebalancing of index ETFs is their most valuable property, rather than their actual choice of equities.

I admire what the Langchain team has been building toward even if people don’t agree with some of their design choices.

The OpenAI api and others are quite raw, and it’s hard as a developer to resist building abstractions on top of it.

Some people are comparing libraries like Langchain to ORMs in this conversation, but I think maybe the better comparison would be web frameworks. Like, yeah the web/HTML/JSON are “just text” too, but you probably don’t want to reinvent a bunch of string and header parsing libraries every time you spin up a new project.

Coming from the JS ecosystem, I imagine a lot of people would like a lighter weight library like Express that handles the boring parts but doesn’t get in the way.

Wow so funny to see this post and comment right now, I’ve been writing out a lot of thoughts/theories on consciousness the last few days, and came to a very similar conclusion as you.

I don’t follow you. I’m saying we haven’t discovered any inorganic consciousness, so it isn’t a given that we will be able to create it with digital computers. Not sure how that breaks the laws of physics.

I agree but even if you take materialism for granted, we’ve yet to uncover the exact biological mechanism. It’s entirely possible that it is unique to carbon-based and/or analog brains.

Yeah don't see how domains would be that much better than IPs but it would be easier to understand. It's already quite common for email senders to use multiple domains when they're concerned about deliverability, and domain reputation is already a factor I believe.

Re: deliverability, I do think incumbents are benefiting from the complexity of email, but I'm not sure I follow your argument about Meta.

Oops you’re right!

You can still do this with classes in typescript:

class Hash extends String {}

https://www.typescriptlang.org/play/?#code/MYGwhgzhAEASkAtoF...

The demo they posted demonstrates how to do it. But I don’t think it’s a generally good solution to the problem, it feels like it solves this specific case where the type is a string hash. I think the evolution of this for other types and objects is more like what the OP article suggests.

I wonder if a more natural solution would be to extend the String class and use that to wrap/guard things:

class Hash extends String {}

compareHash(hash: Hash, input: string)

Here's an example: https://www.typescriptlang.org/play/?#code/MYGwhgzhAEASkAtoF...

Is Google significantly behind? I would say from a technological perspective, they're very closely trailing or even surpassing OpenAI in many ways; they've built a formidable ChatGPT competitor in Gemini—not to mention they have a huge home court advantage with billions of people on Search, Chrome, Android, and G Suite. If you zoom out a bit, it's not really a fair fight between the two. More likely, Apple and Microsoft win against Google because they use/buy OpenAI.

Very exciting, would love to read more about how the architecture of the image generation works. Is it still a diffusion model that has been integrated with a transformer somehow, or an entirely new architecture that is not diffusion based?

I couldn’t quite tell from the announcement, but is there still a separate TTS step, where GPT is generating tones/pitches that are to be used, or is it completely end to end where GPT is generating the output sounds directly?

I haven’t seen anyone mention this yet, but I think the concept here was inspired by all the viral hydraulic press videos on Instagram and TikTok. Here’s a similar video showing random objects and consumer products being crushed in slow motion with similar upbeat music: https://m.youtube.com/watch?v=q9BtYEnrkg4