Hasnain says:

“Currently, the hospital’s wards have 44 babies under 10 days old with weights as low as 2 kilograms (4 pounds), some on life support. Every incubator has at least three premature babies in it, raising the risk of infection. Al-Shair said he fears some will meet the same fate when returned home.

“We treat them now but God knows what the future will be,” he said.”

Hasnain says:

Excited to try this out. Some of the new fuzzing results people posted today are mind blowing.

“Claude 3 Opus, March 4th. This is just a few days old and wow: the vibes on this one are really strong. People I know who evaluate LLMs closely are rating it as the first clear GPT-4 beater. I’ve switched to it as my default model for a bunch of things, most conclusively for code—I’ve had several experiences recently where a complex GPT-4 prompt that produced broken JavaScript gave me a perfect working answer when run through Opus instead (recent example). I also enjoyed Anthropic research engineer Amanda Askell’s detailed breakdown of their system prompt”

Hasnain says:

“From accounts of people on the ground, videos and photographs of deadly episode after episode, plus the resultant mortalities from blocking or smashing the crucial necessities of life, a more likely estimate, in my appraisal, is that at least 200,000 Palestinians must have perished by now and the toll is accelerating by the hour.

Imagine Americans, if this powerful U.S.-made weaponry was fired on the besieged, homeless, trapped people of Philadelphia, do you think that only 30,000 of that city’s 1.5 million people would have been killed?”

Hasnain says:

“DeWitt, writing in his solo book Chile Peppers: A Global History, says TAM became widespread in Texas after its introduction. “It was much milder and larger than the traditional jalapeños, and genes of this mild pepper entered the general jalapeño pool. Cross-breeding caused the gene pool to become overall larger and milder.”

Since I know you’re wondering who the inventors are: the clue is in the name TAM II. The hot (but also not hot) new jalapeño is an invention of Texas A&M University. Yes, Aggies took the spice out of life.”

Hasnain says:

“EU regulators found that Apple’s actions had resulted in users paying “significantly higher prices” for music streaming services.

The iPhone maker charges a 30 per cent fee for all sales through the App Store, a cost the commission said had been passed on to consumers in the form of higher subscription charges.

As part of Monday’s ruling, the commission also banned Apple from blocking apps from offering their services outside the iPhone maker’s iOS software.”

Hasnain says:

“The graph confirms that mutex contention is highest (and therefore RWMutex is most useful!) when the arrival rate is high or the critical section has a long duration.

The graph shows something else interesting. The transition from “no contention” to “high contention” follows a straight line. If some values at the transition are taken and their product calculated, it’s clear that the product is always a constant.”

Hasnain says:

“Seriously, though, there’s one common theme to everything in this article: test your assumptions.

I assumed that exponential growth should apply to the witchcraft population in the Kat books. So I tested that assumption with models, and concluded that this applies… in certain conditions
Unckless’s formulas assumed that the Selection Coefficient is low. Because I violated the assumption, I made a simulation to find how the formula would fail.
I assumed that indexes in SQL databases magically speeds things up. I tested it out, and it slowed things down instead.
I assumed that a genealogy database is a good way to track a person’s descendants. After I analyzed the data, it became clear that it doesn’t. I came up with a new set of assumptions to explain why, but now I’m looking for ways to challenge those new assumptions.
I assume that people want to read about kids’ books, population genetics, and PostgreSQL in the same article. The jury’s still out on this one ;)”

Hasnain says:

“So far, we’ve learned that:

The initial app description is much more important than we thought
Coding is not a straight line, agents can review themselves
LLMs work best when they focus on one problem compared to multiple problems in a single prompt
Verbose logs do miracles
Splitting the codebase into smaller files helps a lot
For a human to be able to fix the code
They must be clearly shown what has been written and the idea behind it
Humans are lazy
It’s hard to get the LLM to think outside the box”

Hasnain says:

“These examples always flow in the same direction. Professor Alvaro Bedoya, the founding director of the Center on Privacy and Technology at the Georgetown University Law Center, wrote “It is a pattern throughout history that surveillance is used against those considered ‘less than’, against the poor man, the person of color, the immigrant, the heretic. It is used to try to stop marginalized people from achieving power.” The same pattern is found in the role of technology in decision systems.

The goal of many automated decision systems is to increase revenues for governments and private companies. When this is applied to health and medicine, the goal is often achieved by denying poor people food or medical care. People often trust computers to be more accurate than humans, in a bias known as automation bias.”

Hasnain says:

Bookmarking for later reference.

“Well, there you have it. A tangible, step-by-step plan to get out of the finicky situation that’s a complex legacy C++ codebase. I have just finished going through that at work on a project, and it’s become much more bearable to work on it now. I have seen coworkers, who previously would not have come within a 10 mile radius of the codebase, now make meaningful contributions. So it feels great.

There are important topics that I wanted to mention but in the end did not, such as the absolute necessity of being able to run the code in a debugger locally, fuzzing, dependency scanning for vulnerabilities, etc. Maybe for the next article!”