“There are things known and there are things unknown and in between are the doors of perception.” — Aldous Huxley

I’m Huxley Westemeier (26’) and welcome to “The Sift,” a weekly opinions column focused on the impacts and implications of new technologies.

______________________________________________________

It’s been two weeks since my last Sift article, and a lot has happened. The public release of gadgets revealed at CES have provided many interesting tidbits to cover. Here are a few highlights:

Nvidia’s new 5090 GPU melted power connectors

Nvidia, a company known for producing graphical processing units or GPUs used in AI server farms and gaming computers, has had a rough month. First, Chinese company DeepSeek announced a ChatGPT competitor that didn’t rely on massive amounts of computational power, causing Nvidia’s stock to decrease by over 18%. Nvidia doesn’t just make chips capable of running AI models. They also offer a variety of consumer-grade computer parts intended for hardcore gamers.

Their latest flagship, the RTX 5090, costs a whopping $1999 and substantially improves graphics performance in games like Cyberpunk 2077. The downside? According to PC World, the heat generated by the GPU has resulted in power connectors melting. It’s a problem that Nvidia claims was supposedly fixed from the previous RTX 4090 model. Replacement power connectors aren’t as expensive as the GPU, with average prices ranging from around $10 to $50 depending on the model. It still maintains impressive performance despite a steady supply of over 575 watts (most non-gaming laptops, like the SPA Lenovo Thinkpad, draw around 50-65 watts). However, it’s still worrying that the connectors might need to be replaced. The RTX 5090 might even become a fire hazard if the issues persist, and we’ll have to see how Nvidia responds to the claims in the coming weeks.

T-Mobile and Elon Musk are promising global connectivity for thousands of smartphone users

Anyone watching the 2025 Super Bowl on Feb. 9 likely sat through T-Mobile’s “A New Era in Connectivity” ad, even if it wasn’t particularly flashy or memorable. T-Mobile is now partnering with Elon Musk’s Starlink to provide internet service to people in remote locations where cellular service might have been unreliable or unavailable. It performs like Apple or Google’s newer Satellite Connectivity features but acts like global WiFi without being limited to low-bandwidth messaging or emergency calling.

It’s free for users on T-Mobile, AT&T, and Verizon until July when it will adopt a subscription model. Customers will need a recent smartphone from the last two years with satellite hardware, with the oldest supported models on iOS and Android being the iPhone 14 and Galaxy S21. I’m worried about the security of the new plan, as T-Mobile has undergone multiple breaches in the last few years, and other companies owned by Elon Musk, such as X, have limited data transparency (remember when Musk asked users to upload private medical photos to his chatbot?).

OpenAI’s o3 model proves to be a worthy Deepseek competitor

Only a week after DeepSeek’s R1 model outperformed OpenAI’s o1 model in nearly every task, OpenAI announced and publicly released the o3-mini model for all users (including those on the free tier). Unlike older models, o3-mini spends a chunk of time “reasoning” before giving the user an output. Sometimes, it takes under 10 seconds to respond (if asking for a simple poem or haiku, for example). In my testing, if prompted with a more challenging task, such as troubleshooting code, it can take up to a minute before generating an output. The “reasoning” process seems to reduce the amount of hallucinations, and in my personal experience, o3-mini appears to be much more accurate than the previous o1 model and is closer to DeepSeek’s R1.

Deepseek itself has undergone immense controversy in the three weeks since its launch, as various sources, including OpenAI, have alleged that DeepSeek might have used ChatGPT’s output to help train the R1 model. We’ll have to wait to see if the claims become a lawsuit. Still, if DeepSeek used OpenAI’s chatbot to help generate training information, it could explain the low costs associated with R1’s development.