Month: March 2024

Enlarge (credit: Xinhua Fu)

On one level, we have fireflies figured out. We know the enzyme they use to make light (called luciferase), as well as the chemicals they use in the light-generating reaction. We know them so well that we've turned them into useful tools for studying other aspects of biology, such that lots of people who have never even seen a firefly have used firefly luciferase in the lab.

But on another level, there's a lot we don't understand. Fireflies clearly exercise a level of control over when they light up, and they do so only in specialized organs. And there's nothing like that organ in other species. So, somehow, fireflies evolved an elaborate light-producing organ, and there's no sign of any potential precursors in related species. Which makes it a bit of a mystery.

Now, a pair of researchers from Wuhan, China, (Xinhua Fu and Xinlei Zhu) have started unraveling what's going on at the level of the genes responsible. And, while they haven't produced a complete picture of how evolution built the fireflies, they've brought us a lot closer.

Enlarge (credit: Getty Images)

On Monday, Anthropic prompt engineer Alex Albert caused a small stir in the AI community when he tweeted about a scenario related to Claude 3 Opus, the largest version of a new large language model launched on Monday. Albert shared a story from internal testing of Opus where the model seemingly demonstrated a type of "metacognition" or self-awareness during a "needle-in-the-haystack" evaluation, leading to both curiosity and skepticism online.

Metacognition in AI refers to the ability of an AI model to monitor or regulate its own internal processes. It's similar to a form of self-awareness, but calling it that is usually seen as too anthropomorphizing, since there is no "self" in this case. Machine-learning experts do not think that current AI models possess a form of self-awareness like humans. Instead, the models produce humanlike output, and that sometimes triggers a perception of self-awareness that seems to imply a deeper form of intelligence behind the curtain.

In the now-viral tweet, Albert described a test to measure Claude's recall ability. It's a relatively standard test in large language model (LLM) testing that involves inserting a target sentence (the "needle") into a large block of text or documents (the "haystack") and asking if the AI model can find the needle. Researchers do this test to see if the large language model can accurately pull information from a very large processing memory (called a context window), which in this case is about 200,000 tokens (fragments of words).

Enlarge (credit: peshkov | iStock / Getty Images Plus)

Bitcoin's price hit a record high Tuesday, surging above $69,000 and notching past its prior peak price of $68,991.85 recorded in 2021.

After bitcoin's price sunk in 2022 amid a cryptocurrency industry crash, bitcoin's comeback began in early 2023, Bloomberg reported. But this week's record price owes a lot to the US Securities and Exchange Commission (SEC) approving spot-bitcoin exchange-traded funds (ETFs) in early January, which funneled more than $7 billion into various bitcoin investment products in less than two months, The New York Times reported.

These ETFs are offered by trusted financial institutions like BlackRock and Fidelity, making it more enticing for the crypto-curious to invest in bitcoin without owning any of the cryptocurrency or understanding any of the associated technology. Instead, they can just buy shares of a collection of digital assets with SEC-required protections against fraud or manipulation.

• 

  The Nothing Phone 2a in black. [credit: Nothing ]

Hot off the embarrassing implosion of its messaging app, the upstart hardware company Nothing is back to making phones again. This time it's a budget device, the "Nothing Phone 2a," which is being sold in Europe for 329 euros and sort of coming to the US for $349 on a "Developer program" with limited carrier support.

Just like the bigger Nothing Phone 2, this has a unique rear design full of lights and faux-mechanical cladding embedded under the clear back. On this model, that clear back is plastic, while the front is glass. There is some aluminum somewhere in the mid-frame, but the sides are plastic.

As usual with Nothing phones, it's very odd to be able to see the screw heads on the back but not access them because of the glued-on rear cover. In a world where everyone, Nothing included, pitches "more sustainable" devices, you could be both good-looking and repairable with a design like this, if you just didn't entomb the screws under a clear cover. The phone only has IP54 dust and water resistance, meaning it can't handle much more than rain, so it's not like the glue construction is doing much. Still, it's nice to see a budget phone with some design work put into it, even if it is this phony faux-repairable design language.

Enlarge (credit: Getty)

Linux reached 4.03 percent of global market share in February, according to data from research firm Statcounter. That takes Linux past the 3 percent milestone it reached in June 2023. While we’re still far from the Year of the Linux Desktop, interest in Linux has somewhat grown lately.

Statcounter says it gets its desktop operating system (OS) usage stats from tracking code installed on over 1.5 million global websites generating over 5 billion monthly page views. The only adjustments the firm says it makes to this data are around removing bot activity and adjusting for Google Chrome prerendering. Note that when Statcounter analyzes desktop OSes, it also includes laptop computers, and Statcounter says it may revise its data within 45 days of publication.

• 

  Global desktop OS adoption, per Statcounter. [credit: Statcounter ]

As spotted by Linuxiac, Linux’s reported desktop market share was higher than ever in February. If you count ChromeOS as a Linux OS, then market share totaled 6.34 percent in February, although that number is actually smaller than what Statcounter reported in June: 2 percent.

Enlarge (credit: Getty Images | Dan Kitwood )

Elon Musk and X Corp. were sued yesterday by four former Twitter executives who say they were cheated out of more than $128 million in severance when Musk bought the social network and fired them.

Former Twitter CEO Parag Agrawal, former CFO Ned Segal, former Chief Legal Officer Vijaya Gadde, and former General Counsel Sean Edgett filed the lawsuit in US District Court for the Northern District of California. They say they are owed one year's salary, stock awards, and health insurance premiums.

"If anyone around Musk had been willing to tell him the truth, he would have learned that his scheme to deny Plaintiffs their contractual severance payments was a pointless effort that would not withstand legal scrutiny," the lawsuit said.

Enlarge / A collage of some of the "interactive environments" generated by Genie from static images or text prompts. (credit: Google DeepMind)

At this point, anyone who follows generative AI is used to tools that can generate passive, consumable content in the form of text, images, video, and audio. Google DeepMind's recently unveiled Genie model (for "GENerative Interactive Environment") does something altogether different, converting images into "interactive, playable environments that can be easily created, stepped into, and explored."

DeepMind's Genie announcement page shows plenty of sample GIFs of simple platform-style games generated from static starting images (children's sketches, real-world photographs, etc.) or even text prompts passed through ImageGen2. And while those slick-looking GIFs gloss over some major current limitations that are discussed in the full research paper, AI researchers are still excited about how Genie's generalizable "foundational world modeling" could help supercharge machine learning going forward.

Under the hood

While Genie's output looks similar at a glance to what might come from a basic 2D game engine, the model doesn't actually draw sprites and code a playable platformer in the same way a human game developer might. Instead, the system treats its starting image (or images) as frames of a video and generates a best guess at what the entire next frame (or frames) should look like when given a specific input.