Black Friday seems to get longer with every passing year, but even when a day becomes a week (or several) it still signals the holiday shopping season is in full gear with Christmas looming a month in the distance. As the trees go up, decorations adorn every facet of our communities, and Christmas music begins to feel inescapable, many people are wondering what to buy for their friends and loved ones — and it’s not uncommon for some those gifts to end up being some hyped-up tech product.

There’s nothing inherently wrong with gifting those you care about a piece of technology, but far too often people aren’t thinking about the broader consequences that come with those gifts — and are then foisted onto people who may be completely oblivious to what they’re signing up for. As givers of gifts, we need to consider what else we’re giving with popular tech products that frequently get passed on this time of year.

As we head into the holiday season, I decided it would be a good time for a guide to help you consider what tech not to give to those you care about. Trying to make people’s lives a bit more convenient and stress-free is great, but sometimes that also means saddling them with more surveillance and potentially even worse as a result. Feel free to share the guide around so others are considering what they’re really giving when they gift certain tech products.

That score is seriously impressive because it actually beats the average human performance of 60.2% and completely changes the narrative that you need massive proprietary models to do abstract reasoning. They used a fine-tuned version of Mistral-NeMo-Minitron-8B and brought the inference cost down to an absurdly cheap level compared to OpenAI's o3 model.

The methodology is really clever because they started by nuking the standard tokenizer and stripping it down to just 64 tokens to stop the model from accidentally merging digits and confusing itself. They also leaned heavily on test-time training where the model fine-tunes itself on the few example pairs of a specific puzzle for a few seconds before trying to solve the test input. For the actual generation they ditched standard sampling for a depth-first search that prunes low-probability paths early so they do not waste compute on obvious dead ends.

The most innovative part of the paper is their Product of Experts selection strategy. Once the model generates a candidate solution they do not just trust it blindly. They take that solution and re-evaluate its probability across different augmentations of the input like rotating the grid or swapping colors. If the solution is actually correct it should look plausible from every perspective so they calculate the geometric mean of those probabilities to filter out hallucinations. It is basically like the model peer reviewing its own work by looking at the problem from different angles to make sure the logic holds up.

What's remarkable is that all of this was done with smart engineering rather than raw compute. You can literally run this tonight on your own machine.

The code is fully open-source: github.com/da-fr/Product-of-Ex…

Hello everyone 🌿

I’m applying to the Foresight Institute — AI for Science program, and I need one neutral reference contact (full name and email) — not a recommendation, not a letter, and no endorsement of the content.

The role is minimal:
If the committee decides to reach out (most likely they won’t), they may ask only:

1. whether you have seen or read the work;
2. whether the application appears serious.

I am developing an interdisciplinary model called ICT (Information–Consciousness–Temporality).

At the core of the model:
— dI/dT as a formal dynamic of consciousness,
— I_fixed as a model of material fixation of informational states.

Discussion and preprint:
academia.edu/s/8924eff666

PDF: academia.edu/144946662/The_Con…

DOI:
zenodo.org/records/17584783
Docx format

If any researchers here are willing to serve as such a neutral contact, I would be very grateful.
It requires zero time from you other than possibly confirming briefly by email.

Thank you to everyone who responds.

Cars used to be entirely mechanical objects. With hard work and expertise, basically any old vehicle could be restored and operated: On YouTube, you can watch a man drive a 1931 Alvis to McDonald’s. But the car itself was stuck in time. If the automaker added a feature to the following year’s model, you just didn’t get it. Things have changed. My Model 3 has few dials or buttons; nearly every feature is routed through the giant central touch screen. It’s not just Tesla: Many new cars—and especially electric cars—are now stuffed with software, receiving over-the-air updates to fix bugs, tweak performance, or add new functionality.

In other words, your car is a lot like an iPhone (so much so that in the auto industry, describing EVs as “smartphones on wheels” has become a go-to cliché.) This has plenty of advantages—the improved navigation, the fart noises—but it also means that your car may become worse because the software is outdated, not because the parts break. Even top-of-the-line phones are destined to become obsolete—still able to perform the basic functions like phone calls and texts, but stuck with an old operating system and failing apps. The same struggle is now coming for cars.

Software-dependent cars are still new enough that it’s unclear how they will age. “It’s becoming the ethos of the industry that everyone’s promising a continually evolving car, and we don’t yet know how they’re going to pull that off,” Sean Tucker, a senior editor at Kelley Blue Book, told me. “Cars last longer than technology does.” The problem with cars as smartphones on wheels is that these two machines live and die on very different timescales. Many Americans trade in their phone every year and less than 30 percent keep an iPhone for longer than three years, but the average car on the road is nearly 13 years old. (Tesla didn’t respond to a request for comment about how its cars age.)