Category: Uncategorized

Is our society increasingly rewarding conformity? Is AI accelerating this process, and is it simultaneously stripping work of deep meaning? After all, thanks to LLMs, many now define creativity as merely a probabilistic recombination of matrices derived from a training set.

Is science also contributing to this potential lack of deviation in culture, education, arts, architecture, and business? Here’s a take on it (not mine):

You can spot this scientific bland-ification right away when you read older scientific writing. As Roger’s Bacon points out, scientific papers used to have style. Now they all sound the same, and they’re all boring.

Maybe science is supposed to be boring? What happened to the style though?

For definitive answers to these questions, look elsewhere. For a compelling set of data (plus plenty of causal speculation) on many aspects of contemporary society and scientific style, check out this compilation and the essay here.

The greatest thinkers in science (and business) are often prolific authors. They write books, blogs, and copious emails to sharpen ideas. Richard Lewontin, E.O. Wilson, and Paul Graham are but three examples. Dorothy Hodgkin’s scientific correspondence and papers, stacked together, extend 25.85 meters in length. Great thinkers, in other words, write all the time.

Researchers are evaluated by simple measures of productivity or influence — number of papers published, citation count, and grant dollars. In such an environment, it has become exceedingly difficult for scientists to take stylistic risks in their academic writing or to devote significant amounts of time to other forms of creative writing.

Here’s an update on using LLMs for OCR without having to use the same hammer (generic model) for all nails. DeepSeek has released an OCR-focused model: https://github.com/deepseek-ai/DeepSeek-OCR

Check out the deep parsing mode, which is parsing images within documents through secondary model calls. Very useful for data extraction. The results are pretty impressive too:

Our work represents an initial exploration into the boundaries of vision-text compression, investigating how many vision tokens are required to decode 𝑁 text tokens. The preliminary results are encouraging: DeepSeek-OCR achieves near-lossless OCR compression at approximately 10× ratios, while 20× compression still retains 60% accuracy. These findings suggest promising directions for future applications, such as implementing optical processing for dialogue histories beyond 𝑘 rounds in multi-turn conversations to achieve 10× compression efficiency.

The data on education call for attention:

– 33% of eighth graders are reading at a level that is “below basic”—meaning that they struggle to follow the order of events in a passage or to even summarize its main idea.

– 40% of fourth graders are below basic in reading, the highest share since 2000.

– In 2024, the average score on the ACT, a popular college-admissions standardized test that is graded on a scale of 1 to 36, was 19.4—the worst average performance since the test was redesigned in 1990.

The article speculates on several causal links to explain the declining trend in the metrics, ranging from the effects of COVID to the influence of smartphones and social media.

The point that truly resonates with me as an educator, though, is this: a pervasive refusal to hold children to high standards. Standards are about values, not technology or tools. No tool causes the fading emphasis on rigor.

The article discusses other important aspects, such as the disparity between school districts, the heterogeneity in outcomes based on affluence, and the potential role of AI as a democratizer, but keeps returning to the same line: declining standards and low expectations. And that’s for a good reason:

“Roughly 40 percent of middle-school teachers work in schools where there are no late penalties for coursework, no zeroes for missing coursework, and unlimited redos of tests.”

This is potentially the most important problem facing our society today, and it warrants far more attention.

Source

It didn’t take long for Tom Fishburne’s cartoon to come true: Generative AI is increasingly treated as a magic trick. Most tricks aren’t really useful or truly “real,” yet they remain entertaining until we figure out, often the hard way, how the illusion is performed.

My conversations show that some executives are cautiously optimistic, integrating only what is truly useful into their business processes and project workstreams, while others are applauding the tricks and wanting more. The former group is already AI-literate, using machine learning and algorithms to augment and automate processes in the broader, more accurate definition of AI. The latter, more easily impressed group, seems to lack this foundation.

We can expect that more of the applauding executives will eventually join the cautiously optimistic ones once the magic show ends and we move past the peak of the hype cycle.

This is an (almost) complete picture of LLMs, highlighting the underlying math for probability calculations. It’s a great companion for the 64-picture visual summary I curated for a talk earlier this year.

Good reminder that “Life’s most important questions are, for the most part, nothing but probability problems.” Pierre-Simon Laplace in Théorie Analytique Des Probabilités

I once quoted Abraham Maslow in my note “Using predictive modeling as a hammer when the nail needs more thinking“, a problem that has since eased as focus shifts to causal modeling and optimization.

Now, if the problem is extracting handwritten digits from a PDF or an image, what’s the solution? I guess good old reliable OCR? There is nothing generative about extraction, so using an LLM may be moot, if not actively detrimental (due to hallucinations). OCR is widely available through cloud like AWS Textract and Google Cloud Vision, some of which also offer human validation.

Instead, the Reuters team intriguingly chose the shiniest hammer, an LLM, to extract data from handwritten prison temperature logs. They uploaded 20,326 pages of logs to Gemini 2.5 Pro, followed by manual cleaning and merging. One big problem with this approach is the inevitable made-up text (hallucinations), which required the team to hand-code 384 logs.

So why not OCR but an LLM? All I can think of is, LLMs may be useful in extraction when the text is highly unstructured and contextual understanding is needed, neither of which seems to apply here. Surprisingly, the methodological note doesn’t even mention OCR as a solution.

Putting the tool choice aside, though, Reuters asks an important question here: “How hot does it get inside prisons?” and the answer is “Very.” I applaud the effort and data-centric journalism, and I recommend reading the story.

Credit for the image goes to Adolfo Arranz.

Source – Project GitHub