Hacker News Reader

baschon I'm Tired of Talking to AI

The writing is on the wall. We are headed for a world where everybody interprets everybody through a personalized model. (corporations too.)

Our models need to understand each other, we don't need to understand each other. A call and response to the tower of babel. We eventually all learn to speak our own custom language known only to us. Our inner monolog moves externally, and we offload "understandability" to an external entity.

baschon Uber president says AI spending is getting 'harder to justify'

It was very easy to see before this started that the models were just repetitive inflators.

It’s not that they should “scale back” their use as much as the metric should be improvement/tokens. Tokens used is a denominator in any worth calculation.

baschon The seed oil panic is hurting my cardiac patients

I remain unconvinced that hexane in the oil poses any serious risk.

It has such a low boiling point, it’s heated. It all evaporates out. What’s the mechanism keeping hidden undetected hexane in the product.

baschon The seed oil panic is hurting my cardiac patients

There is also sunflower oil and high oleic sunflower oil, the latter, which after refinement is incredibly heat and shelf stable, and is essentially pure omega-9 monounsaturated fat.

The anti refinement process perspective is discounting that the end result is the perfect fat.

A similar thing can be said about hydrolyzed collagen with a little tryptophan added.

baschon The seed oil panic is hurting my cardiac patients

>Seed oils probably aren't ideal

Except they are. A refined monounsaturated fat is in the upper tier of perfect calorie. It makes a lot of sense for omega-9's to be a major part of caloric consumption. The omega-3 and omega-6 when fresh and uncooked are also unnecessary demonized by the idea that they are immediately rancid. Like many things the problems arise from the implementation or how they are used, not the chemical itself.

baschon Schlitz Is Gone, but First It's Getting One Last Hurrah

Buried at the end of the article is the real story. "Schlitz" hasn't existed since the early 70's. They cut costs and ruined the formula and brand. In 2008 it was "revived" and a new beer took the name. Now someone else is brewing a completely different reconstructed formula for "the last batch" and throwing the name on it, again. And im sure itll happen again and again after that.

baschon Microsoft starts canceling Claude Code licenses

since those headlines started ive felt it just encouraged inefficiency. "say as much as you can without saying anything." if you were accomplishing your task the need for more would end, thus there is incentive to never succeed.

baschon Google's Antigravity Bait and Switch

gpt-image-2 still has a very visible banding/artifacting, but nb2 and gpti2 each have their strengths to the point that its momentarily worth running the same prompt through both and grabbing the better result, then feeding it to the opposite model to tweak it. both of them do a better job of not "getting stuck" regurgitating the same thing over and over when the base image comes from the opposite model. i tend to try and limit how many edits i make with gpti2 because successive iterations degrade the image much much faster.

huge waste of firefly credits tho.

baschon Google's Antigravity bait and switch

Whisk is a good example. They put out a pretty slick image manipulation tool that had infinite potential, and decided to just dump it and work on Flow instead, which while similar in features has a much more cumbersome interface. They really dont learn what works well about something when shuttering it. Poor post mortem anaylsis. Whisk very much could have lived on as a parallel evolution of interface over their image gen stack. I know it was labs software, and should have been expected but man they love throwing away what works for what doesnt.

baschon Google's Antigravity bait and switch

Which somewhat ironically can import from github but not directly from aistudio..

It's a little crazy they still depend on microsoft as an intermediate between all their tools.

baschon Why I don’t vibe code

The next version from that vendor. Or maybe the market commoditization.

You’re again conflating the product offerings from the capacity of the technology. It’s like kinetic energy vs potential energy.

The real wait is for the cost in disk space, memory and electricity of pretraining a transformer to be something the equivalent of a raspberry pi can accomplish. Then all the style and design choices of the few current vendors goes away.

baschon Why I don’t vibe code

People have a very strong tendency to go "current gen technology cant achieve x, therefore its impossible for this class of technology to ever achieve x"

Many many many of the problems with current gen ai will go away.

baschon Ex-Apple engineer says Apple deliberately slows older phones via updates

well and where is the line between true intent vs lack of caring. if bloat is a side effect of feature creep and project handoffs, slowing old devices is an "unintended" feature.

baschon Remove–AI–Watermarks – CLI and library for removing AI watermarks from images

The distortion is shockingly visible in images. Especially with any amount of generational iteration https://streamable.com/9x3s4r

That said, this tool is incredibly lossy, garbling text, completely changing shapes. It also fails to remove the new gemini spark / placement that moved since yesterday.

baschon Ex-Apple engineer says Apple deliberately slows older phones via updates

or just bloat. make no attempt that your newer versions are as optimized because newer hardware covers your inflation.

baschon Gaussian Splat of a Strawberry

I could see a kind of fun game / design tool / worldbuilding where you get a blurry world and you describe what you are seeing, and it comes into focus. The game world, mechanics, aesthetic, and playstyle build as you form your view. A sort of fog of war meets rorschach game.

baschon Going full AI engineer, not touching code anymore

> If an LLM gives you an immediate, coherent answer, there's a risk you accept the framing too quickly instead of doing the harder work of questioning premises, testing alternatives and noticing what might be missing.

Arguably having a conversation with something so confident but random makes me more inclined to "verify before trust" instead of the usual "trust but verify." Im already very question premisy but interacting with ai 10x's that instinct. My skepticism level is higher because of temperature and seeing when I ask something multiple times I get dice roll answers.

baschon Native all the way, until you need text

Did you look at the app? It's a todo list. The idea of a multi gigabyte todo list makes me uneasy.

baschon The Quiet Renovation at Bitwarden

as long as the people who signed up when it said it are granfathered, is it ok then?

baschon The Global Fertility Crisis Is Worse Than You Probably Think

except automation and technology multiplying labor, right? isnt the point of all this innovation to do more with less. are you saying that fewer people, with the assistance of robots, cant provide food, medicine, shelter, and water to the aging?

baschon The AI water issue is fake

https://www.politico.com/news/2026/05/08/georgia-data-center...

>The developer, Quality Technology Services, owed nearly $150,000 for using more than 29 million gallons of unaccounted-for water.

>The company said its water consumption was so high last year because of temporary construction-related activities, such as concrete work, dust control and site preparation.

I do find it interesting that the framing is "a data center" used a bunch of water if it really is "manufacturing concrete uses a bunch of water."

Different source and event, same misleading headline as the one mentioned in the parent article.

baschon AI is a technology not a product

and the differentiating factor on hardware will be the seamlessness of the interface, in software. the combination of voice, eye tracking, swiping, capture of intent, being able to mumble to myself at a volume only my device can hear. The hardware needs to be little more than something that gets out of the way and acts as an input device with a battery.

baschon Google Cloud fraud defense, the next evolution of reCAPTCHA

Is this why google was repeatedly telling me I was displaying patterns of being a bot yesterday because I click too fast? I've never gotten the error message as many times as I did yesterday.

baschon Show HN: Explore color palettes inspired by 3000 master painter artworks

Is there enough color data left in the brown to correct it?

Or do you need to infer it based on location, budget, time, climate etc?

baschon Accelerating Gemma 4: faster inference with multi-token prediction drafters

I have no idea when I’m being lied to anymore but allegedly Aider and Cursor work the way I described, although cursor is using a second model to apply the edit.

baschon Accelerating Gemma 4: faster inference with multi-token prediction drafters

So predict the tokens of the operational transformation.

I just asked: Write the operational transformation sequence and command to turn “this is really beautiful” to “this is very very beautiful”

and in return got: You can map this out by moving a virtual cursor across the text and telling it what to keep, remove, or add. You start by retaining the first eight characters to keep "this is " untouched. Then you delete the next six characters to remove the word "really". In that exact spot, you insert the nine characters for "very very". You finish the operation by retaining the final ten characters, which preserves the space and the word "beautiful". You can code this specific command sequence as [retain(8), delete(6), insert("very very"), retain(10)].

In a large paragraph of text I would expect it to be way quicker and cheaper to generate “[retain(800), delete(6), insert("very very"), retain(10000)]” than repredict the entire remainder of the unedited text.

baschon Accelerating Gemma 4: faster inference with multi-token prediction drafters

I am saying to directly output the operational transformation instructions as the tokens. You’re essentially telling it to “write the diff” and then applying the patch.

[retain(8), delete(6), insert("very very"), retain(10)]

baschon Accelerating Gemma 4: faster inference with multi-token prediction drafters

I have a dumb performance question.

Why when asking a model to change text in a minor way; are we not asking it to generate the operational transformations necessary to modify the text, and then just executing the ot on the existing text vs reproducing every token? Maybe tools are doing that more than I realize?

baschon Maladaptive Frugality

It’s just a 3% surcharge at restaurants now. Maybe you haven’t experienced it in your area but most non chain restaurants have it by me.

baschon OpenAI, Google, and Microsoft Back Bill to Fund 'AI Literacy' in Schools

Using AI in an intentional way with purpose and direction should be great for practicing thinking.

The right way to teach children to use AI is to teach them to scope, filter, design process, edit, refine. How to ask a question, how to think through steps, how to use language to describe all these things. Each kid has something that can think and respond as fast as they input.

The goal should be to perfect sequence and iteration, not skip to final output.

These skills also should NOT be framed in some kind of "teaching AI" as much as teaching communication and critical thinking and analysis. It is the exact same skills you need to solve problems and interact with humans.