bionhoward

gemini app is really funny because they ship ridiculously complicated features like this before fixing the basic ability to have a chat history with apps activity turned off

Imagine the meetings where they decide to add personal illustrated storybooks before fixing chat histories

Anthropic goes hard on the virtue signalling but I still don’t get how anyone manages to satisfy “Customer may not and must not attempt to (a) access the Services to build a competing product or service”

So I just avoid it and generally think the whole thing isn’t serious, because nobody seems to care enough about the safety implications of building AGI with legal terms which are logically impossible to satisfy to demonstrate appropriate attention to detail (aka, yall are noobs)

This is great, My 2 cents since I see a Boolean:

what if that setting were an enum every operator started with AiMode::Off so we can consciously opt-in, plus then Zed could have more AI Mode levels besides just on or off?

Maybe there could be a level for experienced programmers and another level for vibe coders, and prompts / permissions could be set accordingly?

Yesterday I wasted a few cursor max requests on Gemini 2.5 Pro because it couldn’t wrap its mind around the fact that I was modifying a nested folder (./workspace/crate existed, and I told it to modify that, it kept modifying nonexistent ./crate assuming it was in workspace) even though I kept telling it. o3 just used “ls” a few times and figured it out.

I want to like Gemini in Cursor, for the 1M token context but for some reason the outcomes don’t match the benchmarks (for me)

Could a Mistral acquisition signal Apple realizes open source AI is a major selling proposition for their premium hardware? Seems like it could be a good thing if Mistral “rubs off” on Apple

does claude code have a privacy mode with zero data retention?

What makes you say Google is explicit about the fact they have humans and AIs reading everything? It’s got a confusing multi-layer hierarchy of different privacy policies which hide what’s happening to folks’ conversations behind vague language. They promote it as being free but don’t even link to the privacy policies when they launch stuff, effectively trying to bait noobs into pasting in confidential information

is sending your whole codebase to xAI a good idea?

Citations! Perplexity does the best job with references.

Also labs seems superior to artifacts / canvas because it incorporates research in advance of building apps, enabling it to gather data for the app and check docs

It’s absurdly dumb, agreed

it reads, “competing products or services” — not “training competing AI models” with their outputs

Also, who the fuck wants a codebase they can’t train models on

That’s what everyone acts like the ToS forbids, but the language of the ToS as written is much more (infinitely) broad

Given the Anthropic legal terms forbid competing with them, what are we actually allowed to do with this? Seems confusing what is allowed.

No machine learning work? That would compete.

No writing stuff I would train AI on. Except I own the stuff it writes, but I can’t use it.

Can we build websites with it? What websites don’t compete with Anthropic?

Terminal games? No, Claude code is a terminal game, if you make a terminal game it competes with Claude?

Can their “trust and safety team” humans read everyone’s stuff just to check if we’re competing with LLMs (funny joke) and steal business ideas and use them at Anthropic?

Feels like the dirty secret of AI services is, every possible use case violates the terms, and we just have to accept we’re using something their legal team told us not to use? How is that logically consistent? Any safety concerns? This doesn’t seem like a law Asimov would appreciate.

It would be cool if the set of allowed use cases wasn’t empty. That might make Anthropic seem more intelligent

This is a cool concept, but for comparison, I can’t help but wish there was more comparison between the treatment group and a control group that doesn’t see any universal pretraining data.

It’s good to compare various model sizes and evaluation tasks and random data generators. I just think the paper would more effectively prove its point if it could show models of same sizes which see this random data can learn better from evaluation data later on.

Could even take the initial checkpoint of the model before universal pretraining against the pretrained checkpoint. If the method works, the one that did UP will win.

Maybe I’m way off, I’ll admit I only skimmed it so far. Seems promising, just wishing for some controls.

If moving faster is a most, then open source AI could move faster than closed AI by not needing to be paranoid about privacy and welcoming external contributions

FYI, there’s a correctness issue in the part about correctness contracts: valid_opts = ['A', 'B', 'C'] if v not in valid_sizes:

valid_sizes is undefined

What’s this? Another category of companies forced to compete with Anthropic? It’s almost like it’s impossible to use Claude without breaking the Anthropic legal terms…

I always get blocked in using Claude Code by basic logic

What are you building that doesn’t compete with Anthropic? (Using your brain competes with Anthropic) — major legal risk

How do we justify accepting the lack of privacy on Claude? Is it just for people doing FOSS? You’re cool with them reading your business codebase to verify you aren’t using your brain?

Given it is logically impossible to not compete with general intelligence, and that I expect private github repos to remain private, I feel forced to think Claude Code is a nerd snipe / bad joke / toy

Wrapper companies are winning these days because using the AI services directly is literally dumb, due to poor privacy and explicitly anticompetitive legal terms. Using wrappers lets you isolate yourself from the noncompetes, ensure no model training on your content, and enjoy a significantly better user experience.

Maybe one day OpenAI, Google, and Anthropic will wake up to the fact their lame privacy takes (e.g.

- OpenAI hiding the opt out of model training for even when other people use your custom GPTs

- Gemini’s implicit human review of every conversation, Gemini breaking chat history when you opt out of brain rape,

- Anthropic’s impossible-to-logically-satisfy customer noncompete giving them an excuse to read everyone’s chats to verify they aren’t competing

) and customer noncompetes are retarded,

but until that happens, wrappers are crushing them and riding a tremendous wave of innovation without needing to waste money on the most expensive part which is training

Would rust be a lot better than HCL to replace yaml? Just learning Kubernetes and want to say I hope Kubernetes 2.0 uses a “real” programming language with a decent type system.

A big benefit could be for the infrastructure language to match the developer language. However, knowing software, reinventing something like Kubernetes is a bottomless pit type of task, best off just dealing with it and focusing on the Real Work (TM), right?

imagine these companies making antitrust complaints against one another while both companies have explicitly anticompetitive legal terms that mean users can’t fine tune ai on data they “own”

Assuming attention to detail is one of the best signs people give a fuck about craftsmanship, isn’t the fact the Anthropic legal terms are logically impossible to satisfy a bad sign for their ability to be trusted as careful stewards of ASI?

Not exactly “three laws safe” if we can’t use the thing for work without violating their competitive use prohibition

Who can take em seriously when cursor has privacy mode while copilot implicitly trains AI on copilot interactions with customers’ private repo code unless you get some enterprise tier of service?

The core of GitHub is about version control, including private repos, the fact they would violate that trust relationship just to chase AI capabilities means they’ve totally lost the plot as far as I’m concerned.

Maybe Copilot is a nice toy for open source projects or noobs working on private codebases who don’t take 5 minutes to read their website, but cursor just seems way ahead in terms of basic respect for customers

lol, might have been good to conduct this study BEFORE making that decision

One funny thing about Google summaries is “copy text” merges all of the links into a giant blob which gets interpreted as a single extremely long broken link. Not a great sign for attention to detail if they don’t even copy their own pasta (it’s been like this for months)

FYI, there’s a bug with the bug, it doesn’t move on mobile

Great job! Congrats on shipping, this looks like a big project

It’s also a matter of competition…there are other AI services available today with various privacy policies ranging from no training by default, ability to opt out of training, ability to turn off data retention, or e2e encryption. A lot of workloads (cough, working on private git repos) logically require private AI to make sense

When has Apple’s strategy ever been to be first to follow on the latest bandwagon? I thought it was more about making great products which integrate design and technology. They have a launch cadence and they take their time, is that really the same thing as “failing?”

There are plenty of places you can find rushed AI projects. Perhaps being slower and smoother is itself a differentiator?

MCT oil could be one option because it’s a shorter chain saturated fat with a number of carbons divisible by 2,

I forget the reason this is better for ketogenesis than longer chain triglycerides, Google answers didn’t seem like what I had learned about it.

Avoiding high glycemic index carbs (sugar, dairy, starch) is a big factor. Also, water: beta oxidation of fat to make the ketones, is a hydrolysis reaction

https://en.m.wikipedia.org/wiki/Ketogenesis https://en.m.wikipedia.org/wiki/Beta_oxidation

> Ketone bodies are not obligately produced from fatty acids; rather a meaningful amount of them is synthesized only in a situation of carbohydrate and protein insufficiency, where only fatty acids are readily available as fuel for their production.