vikramkr

Whether or not the specific policy is good my preference is that changes to policy that have been in force for decades happen based on legislation and not the whims of 9 unelected people. We didn't get clear rules made my legislature, we lost an escape valve that allowed our regulatory apparatus to function while the gerontocracy on capital hill spun their wheels and left everything even murkier than it was before.

Cheveron deference is no longer active unfortunately

Not really - it doesn't seem to be much of a leap to be able to transfer that knowledge to new language. Honestly there's not a lot that's fundamentally new under the sun in new programming languages and solving a problem in a different language is just another problem for an llm to solve - obviously what's easy and hard for an llm is unrelated to what's easy and hard for a human so human intuition should be taken with a very large grain of salt, but my intuition is that it's not meaningfully different for the llm to figure out a new language vs figure out a weird 3rd party api to integrate with. For both of those cases you do need to give them access to docs and examples, it's not innate to the model, but even for languages it does know from its training data you still need to direct it on what style guide etc you want for your codebase to keep it consistent with your other code patterns.

The best AI models at this point are perfectly fine handling new languages and frameworks and stuff - as long as you can point the llm to some docs and some example code it's not gonna do noticeably worse than it would at another language.

That's technically true but practically an agent can just save the script file/rerun it/write a tool that lets it call a reply with memory etc.This aspect is a bit more elegant when it's in the execution context, but the core of "you don't need to predefine tools, the agent can dynamically write its own code" is not exactly new - that's pretty much the basis of why Claude code and codex and all the other agent harnesses are so much more effective than old ways if working with llms - they can just write giant incomprehensible bash scripts on the fly to accomplish random things without having pre-defined tools, write state to files, etc

If 2 out of 100 people I know see a broken website, depending on the website, that's fine, that doesn't sound like a big deal. Now, if out of 10 power users, all ten of them see a broken site once every 50 logins? Thats a much bigger deal. 98% can be more than enough or not remotely enough depending on the units involved but there are plenty of cases where it's fine to not support the last 2 internet explorer users and stuff

That sounds like you're just creating an artificial distinction. If you let other engineers merge their code without looking at it too closely nothing makes the AI any different other than pretending it's "your code" even though it's a git commit full of code you didn't write. At that point you're dealing with functionally a junior engineer's code and if you don't have a culture of good code review etc, that's an issue and maybe you shouldn't be using full agentic coding? You're taking a team of people with no practice in doing code review and making that their entire job - that's not going to end well and just saying "it's _your code_" is basically asserting a fantasy and hoping it works out. If you didn't write the code, it's not your code!

That's understating the progress by a lot - many cancers are a lot more survivable now than previously with better chemo/radio/surgery + immunotherapies + car-t etc etc

Point them at for what?

https://www.thermofisher.com/us/en/home/life-science/antibod...

Moving forward, where an original image is not present or available, the Company will ensure that website users are informed that antibody images may have been optimized for presentation and clarity on the website.

wut. Bro if you don't have an original valiation image then the answer is not to say "oh we'll make sure we communicate that we're making up a random image" - it's to say you don't have the damn image. It's validation data wtf. It's not a pretty background image it's validation data if you don't have the data wtf are you "optimizing for presentation?" This faq is unreal - pure CYA except by someone who doesn't seem to know what they're trying to cover. If you've got cut and pasted/rotated bands that's just fake data. Not "optimized for presentation."

Yes labs should and usually do always validate new antibodies as well. It's a waste of time and taxpayer money for them to spend their time on bad antibodies they purchased based on fake validation data. And just fundamentally - don't make up validation data. If it's not there it's not there. What are you optimizing for presentation if there's no original!? What does that say about the rest of your process?

What's wrong with the cloud? I get the point about a hype cycle but those two examples don't seem even remotely in the same universe of similarity. SQL is still around, but the cloud won pretty comprehensively no? If you're starting something new, there's still a debate around what's the best database to use and stuff, but I'm struggling to imagine a content where the idea of managing your own hardware would even make sense as something to consider outside of the context of an existing large company with pre-existing on prem infra? And at least in the web world, 90+% of the time you are just going to go and click some buttons in vercel and posthog and whatever and that's your infra. The generic saas webapps that make up a huge chunk of the new projects that would be making architecture decisions where you might theoretically consider on prem are pretty well covered by the point and click abstractions - most of them probably wouldn't exist with the friction and overhead of having to actually manage bare metal

It's not just about returns, it's also about risk. The role of a passive index fund is to be a passive index fund. If the s&p starts chasing returns, that will reduce its utility to the market. You get higher returns by being compensated for risks that passive investors/retirement funds don't want to take. And active investors use the S&P and similar indexes for the specific risks and asset class exposures they provide. You might think the economy is going to do poorly which would be good news for some company that's anti-correlated with the economy but you need to hedge that bet for if the economy does well, so in addition to buying shares of whatever that company is you buy into some market reflectiong mix of stocks bonds etc. The role of that hedge is to have a counterbalancing asset that moves opposite your primary bet to reduce volatility, and the role of the s&p 500 is to broadly reflect the american large cap publicly traded stock component of the market. If the S&P 500 begins behaving unpredictability to chase returns as an index then buying funds that track that index is no longer doing what you need it to do. S&P index loses utility, active investors just use some other index, but passive investors with 401ks locked in to tracking the s&p are suddenly forced to buy whatever bet the index creator is making en masse driving the stock price up. That's not a good outcome for anyone except the company muscling their way in and anyone that was somehow rewarded by that addition

The post training is meant to make it more steerable (usually). I might not want it to write tests. I might not have a dev environment set up for an agent to run tests in it's loop. A major goal in post training is to make it follow instructions, which doesn't have to mean have particular instincts for code organization

Why is that hard to believe? It's literally the prompt telling it what to do - if you want a poem about watermelons you tell it to write a poem about watermelons, if you want tests you tell it to write tests. It's not like TDD is some universal pattern that every llm will naturally optimize towards

this model is whack. Exclamation marks everywhere, sycophantic - not producing working code on prompts the other models handle fine.

"The reason it is echoing back your messages is because gpt-5.4-nano is a fictional model name!"

"Everything is in perfect order! Let's-Go-ready for the next phase, which will connect this durable infrastructure to the user-facing UI!"

It's like they RLed it on thumbs up and downs on ai overview responses and forgot to make it not be a sycophantic echo chamber machine. And like, the thing it built doesn't work because it's not actually in perfect order, but it doesn't seem to be able to figure out what's wrong because everything is clearly remarkably engineered

That's a list of like 6 things. And each of those less complicated a question then the seven thousand questions people throw at you when you complain about something not working right on a Linux distro or about speeding up build times for a new tool or configuring webpack or like pretty much any software tool. What lint rules are you using are you using poetry or uv are you running on Mac windows linux or wsl how are your security groups configured in aws - some tools are more plug and play but it's quite the stretch to say that asking "how is your code organized, do you have your agents.md config file set up, do you have tests, and how large is the codebase" is some sort of unmanageable list of questions for a software engineer to think through when figuring out wtf is going on with some new tooling they're using

My take is there was one big inflection point around opus 4.5 when they got the agentic stuff working and now whether or not it works depends on whether your use case/area of software engineering is profitable enough for the companies to have spent a bunch of money generating synthetic data to RL on, or if it's similar enough to areas that they've done that for. With similar enough being a very loose constraint given how much overlap there is in a lot of coding fundamentals. Tbh if the models aren't working for you now I don't think they're gonna be working for you in 6 months

It's very real but probably very domain specific. It got really good at a lot of traditional web dev stuff, bash, sql, and writing one off scripts to accomplish random tasks (hence all the agent stuff taking off). And they got good at staying on task. That may not translate to game dev because from what I understand a lot of these gains are basically around post training methods driven by synthetic data generation etc (with potential caveats on how synthetic that data actually is lol). I wouldn't be surprised if the areas of code the llms are good at now are straight up just product decisions of where to allocate budget for generating those synthetic data sets, and game dev stuff might not be at the top of the list because the customer base for that might not be as big

If I had to guess - lumber costs might be dominated by labor costs? If they don't have guest worker programs it might not be cost effective anymore as wages go up

That's not actually a thing. Very few trees we plant have specific male vs female plants. One of the few that does that gets brought up in this context, ginko, tends to have male trees preferred because the fruit kind of reeks. Ginkgo fruit is also toxic so you really don't want masses of it getting washed into local waterways in ecosystems the tree isn't native to - not a great time for the local wildlife. A significant supermajority of all the rest of the trees that you plant in cities are gonna have male and female flowers on the same plant or male and female structures within the same flower.

For examples from more recent history think giant scissors used for ribbon cutting ceremonies, the golden spike used to signal completion of the intercontinental railroad, or like all the stuff related to militaries like changing of the guard that are ceremonial. Or even something like leaving a celebratory emoji on the first or merged by a new hire or a box of donuts on someone's birthday. Or bringing out the champagne after closing on a house. Theres just a ton of ritual in day to day life, and even more surrounding big high impact moments that might leave behind a bunch of artifacts like weddings and funerals

TanStack is no way saver than npm. No one understands TanStack.

Pandas is also in no way safer than pip. Because pandas is a library and pip is a package manager and that comparison makes no sense lmao. It sounds like you maybe don't really get or use typescript and don't even really use like basic mypy style types in python (or don't get the difference between what a zod/pydantic validator does vs what a mypy/typescript type system does - zod is also only on the boundary). Which is OK but but there's a difference between not getting why a stack is useful or not having experience with it versus confidently and comically declaring that nobody else understands types either while seeming not understanding what any of the parts here do

For models that reveal reasoning traces I've seen their inner nature as a word calculator show up as they spend way too many tokens complaining about the typo (and AI code review bots also seem obsessed with typos to the point where in a mid harness a few too many irrelevant typos means the model fixates on them and doesn't catch other errors). I don't know if they've gotten better at that recently but why bother. Plus there's probably something to the model trying to match the user's style (it is auto complete with many extra steps) resulting in sloppier output if you give it a sloppier prompt.

Just a tier list I think

That's only like a 100 dollar student discount right? Seems pretty standard

I'm seeing a ton of comments like that one about how apple is holding ipados back but am I going crazy or wasn't the big story of iPad is last year how many updates it got to make it more of a desktop replacement? Like half the features you mentioned were added last year plus a calculator app right? There was specifically a whole iPad os refresh that was well received as finally massively boosting iPad pro capabilities? Like, very recently?

I mean durability is as or more important than repairability and apple products have a reputation for lasting a long time and holding their value. And getting software support etc. In general I think tech nerds underestimate how much people value durability over repairability and how hard it can be to sell repairability. Ive found that "this product is repairable" can be interpreted by people im trying to convince to buy a framework as "this product will need to be repaired and it's going to be your problem." On the other hand Apple's reputation for durability means that buying a used MacBook to save money is a serious and popular option which is by far the most environmentally friendly option compared to any new device, however repairable. The repairability <-> sustainability relationship isn't as straightforward as people suggest in the real world imo

I haven't been on campus in a few years but even then paper was basically absent on campus. A class where a professor wouldn't allow tablets or laptops to take notes would be an aberration and a PITA. I remember I had to write like a paper check once and I had to physically go buy a pen since neither I nor anyone around me had a regular writing utensil on hand.

The exception was when people were taking orgo or a diagram heavy class. For that semester not everyone would have a tablet and some people would have pens and pencils. Or writing classes that still required a handwritten essay for the final exam

If you live near a community bio lab see if you can join up and take some classes to learn some basic lab techniques. And some sort of intro bio class via mooc/textbook/local college class whatever if you can but community lab is honestly a great place to start if you have one.

The main thing to keep in mind is that all the stuff that involves analogies between software and biology is almost universally a bullshit oversimplification that you can safely ignore. It's just that software is so profitable and there's so much vc money in it that there's a ton of pressure to be like "oh we can program biology like we program computers." We can't - we invented computers but didn't invent biology. Biology is the end result of 4 billion years of unchecked entropy - it's a chaos system, non deterministic in the wildest ways, impossibly complicated, and yet something we are getting astonishingly good at understanding and engineering.

Basically, all the biologists that started companies that were like "we can program biology like we can program computers" are bankrupt now.

On the other hand, the computer scientists that respected the nature of biology and pushed the limits of computing to develop Alphafold - giant models trained on the full complexity of biological data - finally created computer systems that could handle biological systems like protein folding at an extraordinary level of capability. They won a nobel.

I mean - steam deck was a pretty significant inflection point quite recently. Making gaming viable on linux via a popular consumer product is a huge deal and starts to kill one of desktop linux's single biggest barriers to adoption.