bormaj

Slightly off-topic, but can anyone comment on the quality of the EV charging infrastructure in the EU? Outside of owning an EV and charging it at home/work, are fast chargers reliable and abundant enough for road trips?

I recently looked into renting an EV in Spain and instead opted for gas. It seems like the public charging infrastructure is just not all that reliable there (e.g. broken chargers + delivering less power than advertised) and fragmented when it comes to paying across many apps. Maybe this is specific to Spain after all.

Yes, but also this particular company has the means to justify the expense. I think there's enough opportunity at scale in the industry to really change the business landscape.

Aside from that (and assuming a large enough sample size) I think it's a safe experiment to at least bet on finding profitable use cases. In 1-2 years, after this experiment runs its course, not everyone will have "unlimited" usage.

What's different this time around?

The author made compelling points regarding capex cycles and supply/demand imbalances. So how does NVDA continue to deliver these returns over the next 10 years? Are the 4 firms driving >60% of NVDA's revenue going to maintain that high capex?

All it will take is a market shock that forces just one of these companies to pivot modes from spending to cost cutting for NVDA's bottom line to see material compression. This is not to say that NVDA will disappear, but it's a very real risk that sets the stage for contagion given how coupled all this growth is.

Pretty neat suite of visuals there! What's the helix intended to represent?

By the way, this post was totally built with the help of another language model which is not Claude!

Was an LLM really necessary to write this article? The LLM fluff glosses over a few interesting points/questions:

- Is the current claude debacle the result of compute supply that cannot match the demand? Or is it that Anthropic is being pressed to monetize its work?

- If LLM inference quality is downshifted because of those constraints, that becomes a very real problem for the end users as it effectively amounts to rug-pulling the product. It is plausible that this eventually materializes itself into the consumer/enterprise plans. E.g. paying up for "high IQ" models and not just token usage.

- As another comment here noted, this is all driving the commodification + privatization of intelligence. The competitive landscape is changing and the effectiveness of "high IQ" models and those that can afford them may very well be table stakes going forward thanks to the (manufactured?) scarcity.

As someone who hasn't yet jumped into working with multiple agents simultaneously, where does a tool like gastown help you the most?

The exception rather than the norm, yeah but IMO his takes are refreshing/insightful.

Gappy is one of the more decorated, public figures in the space. That PDF gives a candid overview of what it's like to interview/work in the industry.

Custom schema wrapper or some package you'd recommend from pypi?

Interesting article, but about halfway through what was a convincing analysis of an ongoing diesel shortage turned into a fear piece on the US/China relations and a corresponding economic collapse.

Diesel production capacity is limited and shrinking. Demand is there, albeit sticky (due to certain demand sources that are not easily electronified) and shrinking at a slower pace (due to slowing global economic growth). Rather than discussing a hypothetical blockade of Chinese oil imports, I wish the author would have addressed more probable short term outcomes re: supply/demand shock triggers and possible market plays.

Feel like this was a missed opportunity to look at recent geopolitical events under a more nuanced lens.

Start small and invest where you have an edge. No edge? Buy the market and keep some gunpowder on the side when an opportunity comes up.

Small players have advantages the big guys don't have at scale and vice versa. The same goes for many markets beyond just finance.

My use case isn't IOT, but about once a month I get a massive data dump from a vendor. Think tens of millions of rows and 100+ columns. Cleaning, ingesting and querying this data via standard RDBMS is a slow and brittle process. There is a time series aspect, but partitioning across other keys/groups is critical.

Exciting project and definitely something I'd like to explore using. I particularly like the look of the API ergonomics. A few questions:

- is the schema inferred from the data? - can/does the schema evolve? - are custom partitions supported? - is there a roadmap for future features?

I don't think Spotify and streaming killed the music subculture, it's still very much alive but requires more intention to find.

Back in the day how did you find new music? Pre-2000s it was likely MTV/radio for mainstream, or word-of-mouth/local events for niche genres. Nowadays Spotify and streaming services have supplanted the former for mainstream music. Finding new music outside the recommended engines requires a little more effort in knowing where to look. There are a lot of Internet radio programs (shout out to The Lot and Rinse.FM) and smaller record labels that do an amazing job at curating local and diverse sounds.

These days it's never been easier to start your own label or publish a track. Rock-'n'-roll is probably still alive (unfortunately I don't know that modern scene well), but assembling the necessary equipment and people to start a band is a big hurdle requiring practice, space and coordination. So I can see more wanna-be artists opting for pop/electronic having shorter turnarounds to a finished product.

I'd love to see a production grade release of this package with async and BCP support. MSSQL has always been a second class citizen (and rightfully so) amongst its open source peers and now it's nice to see MS dedicating resources to this project. They've got some catching up to do, but the alpha benchmarks look quite promising so far!

I think the Catholic church's actions could be rationalized through the lens of preserving influence. Descartes' ideas could be viewed as fostering a mindset of self-reliance and independence which is antithetical to a dependence on the church.

This is a great resource, thank you for putting this together

In other more heavily regulated industries, whistleblowers are fortunately compensated and protected for raising such ethical issues. I wonder how far tech can go before we start to see similar government agencies and rules put in place to do the same.

What's the net benefit to the taxpayer/government for divesting from Fannie/Freddie?

This whole comment is a bit too sensational for me, however I want to highlight one fragment:

When you have so many people alive, who can do nothing in exchange for food [...]

AI, for now, only exists in the digital space and that's what it will primarily disrupt (at least initially). You're still going to need real people to mow the grass, maintain homes/buildings, ship goods and perform other basic services that underpin society. None of that changes drastically with AI in the picture.

Some people will be out of jobs or replaced, but fundamentally they will just have to do something more tangible to provide value in order to pay the bills.

Yeah I see what you mean, you can always disable specific features, but I think that's a habit mypy tries to enforce. They consider redefining a variable bad practice.

Even in rust you have to be explicit about doing the same thing with an extra "let" statement.

If using python with type annotations, linters like ruff and mypy do a great job at identifying issues. It's no substitute for tests and nor will it give you the same guarantees that rust will at compile time. But I think it improves the base quality of the code.

Adding to this, stochastic calculus matters more for modeling volatility/interest rates/derivatives. As you mention, Python/ML are more than suitable for many other areas within quant finance like optimization, algo development, signal research, etc.

I have a couple use cases in mind and am excited to give it a whirl. Thanks for creating this!

This is a really exciting development. Can anyone currently using this chime in on working with pyodide/dependency management in this setting?

I imagine that this reduces the iteration time for developing excel integrations. It's unfortunate that direct db queries aren't supported, but I guess that's a wasm/pyodide issue.

One of the unique aspects of dark pools (and this can vary by venue) is that orders are segmented according to the source type (e.g. retail, client of the operator of the venue, institutional trader, etc) and the quality/grading of their order flow. As an institutional trader trying to leg into/out of a large position, you'll want many different execution options to help minimize price impact + information leakage + execution time. Dark pools are a great option in which you may have the option to specifically trade against non-toxic order flow (e.g. retail) to achieve better execution.

There are other reasons as well, but these venues essentially exist to solve some very niche problems for institutional trading. Public filings are available for review in case you're interested in the rules of each dark pool. [0]

[0](https://www.sec.gov/about/divisions-offices/division-trading...)

HFTs/market makers are competing for pennies on the dollar. They do not hold risk (long term), they exist to readily provide liquidity to your buy/sell orders. HFTs can be quite profitable, but that doesn't make them inherently evil.

Can you elaborate why those bans will improve the market and what the issues are with HFTs in your eyes?

Any specific pitfalls to avoid with K8s? I've used it to some degree of success in a production environment, but I keep deployments relatively simple.

As someone else in the industry, is it fair to categorize all HFs this way?

I think it's fair to say that the context in which a work was created adds to the novelty and ingenuity of it's existence. These works don't exist in a vacuum and there's certainly a difference between a symphony created by Beethoven in his time/setting and a symphony produced as some model's output.

Sure they may functionally have the same effect or enjoyment, but appreciation of a fine work goes deeper than its function.