Ross Ulbricht Pardoned
Amongst the many individuals President Trump pardoned in his first days in office was Ross Ulbricht. Ross is the creator of the Silk Road, a dark web marketplace that was infamous for selling drugs and other illegal items. Whilst it wasn’t the first dark web marketplace to exist, it became incredibly popular by being the first to use crypto as its only method of payment.
The silk road served over 100,000 buyers and made almost $200 million in sales with $13 million in commission mostly going to Ross, which is worth significantly more today thanks to the appreciation of Bitcoin.
Ross was arrested in 2013, and he was convicted in 2015 of double life in prison plus 40 years, without the possibility of parole. At the time,many thought this sentence was outrageous, particularly those who held free market, libertarian beliefs. At the Libertarian Party National Convention last year, President Trump had promised to pardon Ross, and it seems that he has kept his word.
This pardon raises the question of the limits we should impose as a society on free markets, and the significant power that the presidential pardon represents, which has been leveraged by the past two presidents.
DeepSeek
The hedge fund turned AI startup DeekSeek, has taken the world by storm with their new reasoning LLM release, the R1. The founder, Liang Wenfeng had previously founded a quantitative hedge fund, High-Flyer, which had several billions of dollars of assets under management. Conscious of the potential of quantitative research and analysis in finance, they had acquired a considerable number of high-end Nvidia GPUs, which became incredibly useful to train AI models.
The main highlight of this model is not necessarily its performance, in fact, whilst it ranks well amongst frontier models on charts, it is not particularly more performant. Instead, the main advantage is its efficiency. This materialises in lower training costs, claimed to be $5.6 million, and its API costs are 1/30th of OpenAI’s equivalent models.
This has serious implications on the whole AI industry. Not only does it question the capital expenditure of leading AI companies, it also no doubt will open many use-cases previously not financially viable. Nvidia’s market cap dropped $600 billion shortly after the announcement, though it has somewhat recovered since.
DeepSeek R1, which is an open-weight model, came with a detailed technical paper as well. One interesting method of achieving such efficiency was distillation, which can be best described as a master teaching an apprentice. In this case, OpenAI’s reasoning models were asked questions, and the response, which is synthetic data, was fed to train DeepSeek’s model. This is useful as it allows one model to copy the outputs of another model, with fairly limited understanding. Meanwhile, OpenAI is rather displeased with this, though it is hard to pinpoint any wrongdoing. It is also likely to be less powerful than many think, given it is likely that other AI companies use this to some extent as well.
Other ideas employed by DeepSeek include “mixture of experts”, where parts of the model specialise in a certain area, for example coding. When a prompt is given to the LLM, only the relevant part of the model is used, which leads to faster inference.
The widespread hype has caused it to surge to the most downloaded apps on Apple’s app store, and some lawmakers are increasingly worried about the geopolitical implications. Italy has banned it, and members of the US government have warned of the importance of US supremacy in the AI race. Meanwhile, Microsoft (and others) are looking to integrate this model into their product offerings, whilst many others will follow suit.
Apple Earnings
On Friday, Apple shared their Q4 2024 earnings report, where they beat analyst expectations with $124.2 billion of revenue. Despite decreasing smartphone sales, Apple made almost $100 billion last year from services, as part of their push to pivot away from selling physical products to services.
Most notably, Apple is facing increased competition in China, and consumers are generally using their phones for longer before replacing them. This no doubt hit their iPhone sales. However, they also claimed sales in regions where Apple Intelligence is active are greater than areas where not, but it is quite possible that it simply means China, where Apple Intelligence is not yet active, had lesser sales - something we already knew.
Unlike other tech companies, Apple’s stock rose with the news of DeepSeek, highlighting their plan on implementing AI for the consumer. Despite this though, it remains to be seen if customers will willingly pay a subscription for AI features, given people are used to it being free, and are increasingly hesitant for more subscriptions.
Intel’s Blog Post
Intel’s senior management recently released a blog post titled Modular PC Design on their community site, outlining their vision for increased modularity in both laptops and desktops. The post emphasizes the importance of repairability and reducing e-waste, a noble goal but not necessarily convincing coming from intel.
The blog post highlights the importance of making RAM, SSDs, and WiFi cards replaceable, positioning this as a step toward improved modularity. However, these features were simply ordinary practice for decades before recent industry trends moved toward soldered components.
Intel themselves moved to integrate RAM onto a SoC in its latest Lunar Lake processors. Curiously, Intel claims that Lunar Lake was a one-time experiment and that future designs will revert to a traditional CPU-only chip. While this shift might be framed as a commitment to modularity, there are likely other more practical and financial considerations, given the additional cost manufacturing a SoC requires.
On the desktop side, Intel suggests a replaceable GPU, RAM, and CPU, which again, is industry standard. While Intel’s blog frames this as an innovative step, it largely reiterates existing practices rather than presenting groundbreaking changes, and perhaps serves as a distraction from their recent struggles.
Tune in to our podcast available on Spotify, YouTube, and Apple Podcasts for our full discussion on these topics, and talk to us on X at @the_wafflebyte.