weekend ai reads for 2025-02-07

Mistral Small 3 has far fewer layers than competing models, substantially reducing the time per forward pass. At over 81% accuracy on MMLU and 150 tokens/s latency, Mistral Small is currently the most efficient model of its category.

After sifting their dataset of 56K examples down to just the best 1K, they found that the core 1K is all that’s needed to achieve o1-preview performance on a 32B model. Adding data didn’t raise performance at all.

The 1.58bit quantization should fit in 160GB of VRAM for fast inference (2x H100 80GB), with it attaining around 140 tokens per second for throughput and 14 tokens/s for single user inference. You don't need VRAM (GPU) to run 1.58bit R1, just 20GB of RAM (CPU) will work however it may be slow.

It’s not enough for a few individuals to be ahead of the curve in adopting the new skills. Bessen explains that “what matters to a mill, an industry, and to society generally is not how long it takes to train an individual worker but what it takes to create a stable, trained workforce”. Today, every company that is going to be touched by this revolution (which is to say, every company) needs to put its shoulder to the wheel. We need an AI-literate workforce. What is programming, after all, but the way that humans get computers to do our bidding? The fact that “programming” is getting closer and closer to human language, that our machines can understand us rather than us having to speak to them in their native tongue of 0s and 1s, or some specialized programming language pidgin, should be cause for celebration.

These automated agents “read the news” in every town where Good Daily operates, curate the most relevant stories, summarize them, edit and approve the copy, format it into a newsletter, and publish. “Local news should be local. The problem is, at this point, there are economic challenges keeping that from happening. Smaller communities rarely can support enough staff to run a traditional news organization,” said Henderson, who currently runs Good Daily from New York City. “I see technology, and LLMs specifically, as our best shot to fix this.”

For instance, when searching for “how do you turn off [adjective] Google AI results,” a variety of curse word adjectives reliably disabled the AI Overviews, while adjectives like “dumb” or “lousy” did not. Inserting curse words randomly at any point in the search query seems to have a similar effect.

Indigenous people make up less than 0.005% of the tech workforce in the U.S., hold only 0.4% of bachelor’s degrees in computer science every year and have one board member at the top 200 tech companies. In 2022, Native-founded companies only received a mere 0.02% of total venture capital funding.

For instance, Gates noted that one of the “sins of computing” is its ability to reinforce false narratives. AI, he acknowledged, could either exacerbate or mitigate this problem, depending on how it is deployed.

Edith Wheeler Memorial Library in Monroe, CT, told me. “If you're going to say, ‘we have 15,000 ebooks on our platform,’ and 5,000 of those are low quality, AI generated or stuff that's just put on there without any kind of like oversight or selection criteria being followed, what are you actually offering to us?”

Previously, every error message used to teach me something. Now? The solution appears magically, and I learn nothing. The dopamine hit of instant answers has replaced the satisfaction of genuine understanding.

LLMs come in many forms, but for this project, I focused on image-to-text and multi-modal models. Both types of models can analyze images and generate text, either by describing images or answering questions about them.

…

In short, the vision encoder sees the image, while the language encoder describes it.

C-suite leaders participating in our survey are more than twice as likely to say employee readiness is a barrier to adoption as they are to blame their own role. But as previously noted, employees indicate that they are quite ready.

When it comes to generative AI strategies, the stakes are high—and so is the uncertainty. These three-year scenarios can help organizations plan their gen AI strategies.

“It actually empowers me, and gives me the confidence to go tackle more and more ambitious things,” he continued. “It’s going to empower you; it’s going to make you feel confident. I feel more empowered today, more confident to learn something today.”

[His] team of Stasi programmers at DOGE are feeding highly sensitive data from the Department of Education through artificial intelligence tools, according to an extremely predictable, yet still shocking, report from the Washington Post.

Nomic’s official Hugging Face connector lets you import, explore, and curate any of these datasets in Hugging Face with only a few clicks. This makes it easy for anyone to see what's in these datasets, to create embeddings from them, and to search and organize these massive and important datasets in new ways.

The demo will be live from Feb 3, 2025 to Feb 10, 2025. It includes a feedback form where you can contact us to report any successful jailbreaks as well as information on our Responsible Disclosure Policy, which we ask that participants follow. We’ll announce any successes and the general results of the demo in an update to this post.

For example, you can write =PROMPT(A1, “Extract all person names mentioned in the text.”) in a cell’s formula and drag the cell to apply the prompt to many rows. Cellm is useful when you want to use AI for repetitive tasks that would normally require copy-pasting data in and out of a chat window many times.

Pickle captures your best look with AI, creating a digital double that's always Zoom-ready.

Measure & improve your mental health in real time with your personal AI chat bot. No sign up. Available 24/7. Daily insights just for you!

I want to live in a community where everyone cares.