rockinghigh

I wouldn't say R-trees solve the problem better. Joining multiple spatial dataset indexed with r-trees is more complex as the nodes are dynamic and data dependent. Neighborhood search is also more complicated because parent nodes overlap.

That's not true when tiling the Earth though. You need 12 pentagons to close the shape on every zoom level, you can't tile the Earth with just hexagons. That's also why footballs stitch together pentagons and hexagons.

Instagram uses it as their main backend. They have hundreds of million of daily users. Some of the critical backend services are in C++.

Tariffs were a huge point of debate in his first administration. The government had to pay $30 billion to farmers to offset the impact of tariffs.

> China implemented retaliatory tariffs equivalent to the $34 billion tariff imposed on it by the U.S. In July 2018, the Trump administration announced it would use a Great Depression-era program, the Commodity Credit Corporation (CCC), to pay farmers up to $12 billion, increasing the transfers to farmers to $ 28 billion in May 2019. The USDA estimated that aid payments constituted more than one-third of total farm income in 2019 and 2020.

That's definitely happening. The US does this through massive government spending on American solutions. The EU is only starting to go that route as well.

Completely agreed. Apple seriously regressed the multi-lingual experience. They probably have a model per language. If you have to mix languages in a sentence, well, good luck!

They add new data to the existing base model via continuous pre-training. You save on pre-training, the next token prediction task, but still have to re-run mid and post training stages like context length extension, supervised fine tuning, reinforcement learning, safety alignment ...

As you suggested, promotions tend to come more slowly. You're also likely to hit a lower ceiling than someone who is better at promoting their work.

You should read the transcript. He's including 2025 in the age of scaling.

> Maybe here’s another way to put it. Up until 2020, from 2012 to 2020, it was the age of research. Now, from 2020 to 2025, it was the age of scaling—maybe plus or minus, let’s add error bars to those years—because people say, “This is amazing. You’ve got to scale more. Keep scaling.” The one word: scaling.

> But now the scale is so big. Is the belief really, “Oh, it’s so big, but if you had 100x more, everything would be so different?” It would be different, for sure. But is the belief that if you just 100x the scale, everything would be transformed? I don’t think that’s true. So it’s back to the age of research again, just with big computers.

France does not reimburse homeopathic treatments anymore. The NHS in the UK went even further, they funded homeopathic hospitals like the Royal London Homeopathic Hospital and reimbursed homeopathy until 2017.

Wasn't the original LLaMA developed by FAIR Paris?

First, the migration to 2.0 in 20219 to add eager mode support was horribly painful. Then, starting around 2.7, backward compatibility kept being broken. Not being able to load previously trained models with a new version of the library is wildly painful.

Selling yourself and your work is always a critical skill. That's how you get hired in the first place.

The vocabulary size is fairly small (128,256) for a multilingual model. I would guess it doesn't require many additional parameters to support these 5 languages as many tokens can be shared.

You don't need much background in Economics to understand that blanket tariffs hurt companies that rely on imports and exports and lead to higher prices. They used an LLM to come up with a formula that's just the trade deficit when you simplify it and called it "tariffs charged to the USA".

They could provide access to the training code. It's useful for training smaller models or distilling larger ones. They don't need to release every details involved in tuning the optimization parameters during the pre-training stage.

I don't believe OpenAI provides on-premise serving.

Getting only 30% of the advertised range would be very unlikely unless you always drive uphill and in the cold. I've had a Model 3 and a Model Y, and would get at least 80-85% of the advertised range in average California weather.

I worked in the pricing team at the time. There was essentially no personalization. The biggest factors were your current location, where you're going, the time of the day, current demand level around you, and current supply of drivers around you. Promotions were slightly more personalized but battery level was not used.

Probably a combination of being a subscription model with auto-renewal and people regretting wasting money for this type of content.

> Results showed increasing the CO2 concentration to just 800 ppm, a level identified as well ventilated, resulted in an increase in viral aerostability. After 40 minutes, when compared to clean air, around 10 times as much virus remained infectious when the air has a CO2 concentration similar to that of a crowded room (3,000 ppm).

> Dr Haddrell said: “This relationship sheds important light on why super spreader events may occur under certain conditions. The high pH of exhaled droplets containing the SARS-CoV-2 virus is likely a major driver of the loss of infectiousness. CO2 behaves as an acid when it interacts with droplets. This causes the pH of the droplets to become less alkaline, resulting in the virus within them being inactivated at a slower rate.

https://www.bristol.ac.uk/cabot/news/2024/virus.html

Ambient carbon dioxide concentration correlates with SARS-CoV-2 aerostability and infection risk

https://www.nature.com/articles/s41467-024-47777-5

The SentencePiece library also implements Byte-pair-encoding. That's what the LLaMA models use and the original Mistral models were essentially a copy of LLaMA2.

The actual paper: Brainwide silencing of prion protein by AAV-mediated delivery of an engineered compact epigenetic editor https://www.science.org/doi/10.1126/science.ado7082

I only look at the details every few months and rank by descending amount to focus on high ticket items. I have certainly missed some of these subscriptions that only happen once a year. If you have enough savings, small expenses are not always worth chasing down.

Canceling SiriusXM is pretty easy via phone. Adobe makes the process confusing to make you believe you canceled even you haven't yet. It also charges high cancellation fees and has threatening wording.

No need for personal attacks. The states and the federal government could better coordinate tax filing, it would not be against the US constitution.

Most homes have access to 240V outlets that charge much faster, at a rate of 20 to 30 miles of range per hour.

They probably mean affogato and espresso.

These models may be helpful if they speed up convergence when fine tuned on business-specific time series.