Yeah, it's counterintuitive because it's a lot more work for a human to draw a picture (much less a photorealistic picture) than to write a few words, but human language grammar actually has a lot of strict rules that makes that stream of letters work as "valid" output, much less "decent" output that kinda matches the prompt/description. Transpose a pair of letters or even substitute a single letter (or token) and you've got an output that just doesn't work, in a way that generated images don't have to worry about.

Yeah, the smarter way to use LLM-based agents is carefully defined tasks. Mozilla describes their vulnerability assessment processes in this blog post.

Mozilla describes the process they've used: building a harness that instructs a model to find a specific category of vulnerability on a specific interface, and then write up its findings. It's a narrow enough context that the model gets specific instructions, and a simple definition of success, and it sets up many such tasks that can be fed into the existing process for verifying and triaging bugs. Note that the output for this LLM pipeline basically feeds into the same interface for accepting bug reports from the public, or from their human contributors within the project.

There's a couple of takeaways here, too:

• This pipeline is model agnostic. Mozilla set it up before Mythos was released, and its description of other models (Opus 4.7, Codex) confirms that Mythos is better but not a true game changer. The ability to swap out other models provides some assurance that the work done to develop the pipeline will be useful when cheaper or better models come along, or when a model becomes unavailable (like when a provider decides a particular model is too expensive to run, or a provider goes under).
• The increase in automated output (and presumably automation-assisted contributions from the public) has given the humans more work to do. Automation in this context actually increases the demand for human labor.
• Other projects will need to develop their own custom pipelines, specific to their project, to get good results from LLM based agents.

There are ways to use these tools, but none of it really seems like a truly revolutionary/disruptive change to how large projects are managed.

I think AI will be profitable for the next generation of AI business models that emerge from the abandonment of the current business model of developing the frontier. But the prerequisite is that the companies give up on developing the frontier and decide that the models they have are good enough, then get hardware optimized for inference on those models, stagnating into long term commodity infrastructure, like providing phone service or electricity for profit.

So yeah, I think many of these technologies are here to stay, but the growth will stagnate this year as data center construction swallows up companies that overextended.

Yeah that was the real problem with hot swappable discs. The kids who actually physically performed that operation were often too impatient and careless to put the disc somewhere safe from scratching, because they wanted to immediately get back to playing.

Back in 2002? I don't think they separated generation and delivery for most utilities, at least in the US. In 1996, federal regulators made it mandatory for utilities with delivery infrastructure to accept generators' electricity on fair/nondiscriminatory terms, and gave them some time to implement policies. Then, the actual generators started negotiating deals, but the early days were a bit chaotic, with issues in California with rolling blackouts, then the Enron bankruptcy, and then generators actually entering long term contracts with some price stability in the early 2000's.

For a typical residential customer who didn't go out of their way to look for side deals with generators, they wouldn't have needed to see their bills be segmented out into generation and delivery, since most of the utilities still already had long term contracts (or owned their own generation facilities) still in effect from before the regulatory reform.

Personally, I didn't see those numbers separated out on my bill until around 2009. And I remember my electric bill in 2000-2005 being roughly 10 cents per kwh, flat rate.

Why does this image look like an AI-generated screenshot? The letter spacing and weights are all wrong.

He's written up his findings in English, for anyone who prefers English over German or text over video.

But basically the JBIG2 image compression algorithm used in those scanners looked for certain repeating patterns, and incorrectly compressed certain portions of the image into "close enough" blocks of pixels. Unfortunately, that meant that scanned number data wasn't guaranteed to be accurate, even when the decoded output clearly looked like a number with no distortion or noise.

It's worth the full read.

• Existing JPEG files (which are the vast, vast majority of images currently on the web and in people's own libraries/catalogs) can be losslessly compressed even further with zero loss of quality. This alone means that there's benefits to adoption, if nothing else for archival and serving old stuff.
• JPEG XL encoding and decoding is much, much faster than pretty much any other format.
• The format works for both lossy and lossless compression, depending on the use case and need. Photographs can be encoded in a lossy way much more efficiently than JPEG and things like screenshots can be losslessly encoded more efficiently than PNG.
• The format anticipates being useful for both screen and prints. Webp, HEIF, and AVIF are all optimized for screen resolutions, and fail at truly high resolution uses appropriate for prints. The JPEG XL format isn't ready to replace camera RAW files, but there's room in the spec to accommodate that use case, too.

It's great and should be adopted everywhere, to replace every raster format from JPEG photographs to animated GIFs (or the more modern live photos format with full color depth in moving pictures) to PNGs to scanned TIFFs with zero compression/loss.

They always win, unless they don't. History is littered with examples of the freer standard losing to the more proprietary standard, with plenty of examples going the other way, too.

Openness is an advantage in some cases, but tight control can be an advantage in some other cases.

I disagree with your premise. The 111th Congress got a lot done. Here's a list of major legislation.

• Lily Ledbetter Act made it easier to recover for employment discrimination, and explicitly overruled a Supreme Court case making it harder to recover back pay.
• The ARRA was a huge relief bill for the financial crisis, one of the largest bills of all time.
• The Credit CARD Act changed a bunch of consumer protection for credit card borrowers.
• Dodd Frank was groundbreaking, the biggest financial reform bill since probably the Great Depression, and created the Consumer Finance Protection Bureau, probably one of the most important pro-consumer agencies in the federal government today.
• School lunch reforms (why the right now hates Michelle Obama)
• Children's Health Insurance Program (CHIP or SCHIP): healthcare coverage, independent of Obamacare, for all children under 18.
• Obamacare itself, which also includes comprehensive student loan reform too.

That's a big accomplishment list for 2 years, plus some smaller accomplishments like some tobacco reform, some other reforms relating to different agencies and programs.

Plus that doesn't include the administrative regulations and decisions the administrative agencies passed (things like Net Neutrality), even though those generally only last as long as the next president would want to keep them (see, again, Net Neutrality).

The agency’s manager sent me a background memo about the woman I’d be playing, a purported 21-year-old university student blessed with physical proportions that are in vogue these days.

In vogue these days? That just reminds me of how every generation thinks they invented sex. Or the Simpsons quote where Mr. Burns describes a past encounter: "We expressed our love physically, as was the style at the time."

Our heads are just loaded with sensory capabilities that are more than just the two eyes. Our proprioception, balance, and mental mapping allows us to move our heads around and take in visual data from almost any direction at a glance, and then internally model that three dimensional space as the universe around us. Meanwhile, our ears can process direction finding for sounds and synthesize that information with our visual processing.

Meanwhile, the tactile feedback of the steering wheel, vibration of the actual car (felt by the body and heard by the ears), give us plenty of sensory information for understanding our speed, acceleration, and the mechanical condition of the car. The squeal of tires, the screech of brakes, and the indicators on our dash are all part of the information we use to understand how we're driving.

Much of it is trained through experience. But the fact is, I can tell when I have a flat tire or when I'm hydroplaning even if I can't see the tires. I can feel inclines or declines that affect my speed or lateral movement even when there aren't easy visual indicators, like at night.