SamDc73

Honestly this reads like a badly-written propaganda that relies on no messured statistic.

From okla's Speedtest Global Index:

Fixed broadband (median download): US: #9 306.86 Mbps Switzerland: #10 300.70 Mbps

Mobile: US: #11 205.71 Mbps Switzerland: #42 118.93 Mbps

the "25 Gbit" is top tier prodcut that while avablie in the US (sure not the whole country) it's not sometinhg the avg in both coutnries actully buy/use...

why Switzerland has such fast internet at a reasonable price

also on prices the per 2023 (couldn't find any newer resoruces) [2] 100 Mbps in the US seems a bit cheaper in the US than swizerland (but the diffrnece is honestly neligable to begin with)

Every home gets a dedicated 4-strand fiber line. Point-to-Point. Not shared. Not split 32 ways.

form a quick serach, this seems to simply be not true, FTTP coverage is 50% in swizerland (comapred to 69% in europe) and 27% in rural areas (compared to 58% in rural europe) per 2024 [3]

The United States prides itself on free markets

boradcam is one of the most regualted/restricted markets in teh U.S (for better or worse) so it really isn't really free market ...

Local governments grant the cable/telecom franchise and levy franchise fees up to 5% of gross revenue (Cable Act 1984, 47 U.S.C. §542). Entrants also need government-controlled pole and right-of-way access; pole/permit fights are what stalled Google Fiber. And ~16–17 states legally restrict or ban municipal broadband.

Section 301 China tariffs (10–25%) since 2018 (Trump) hit optical fiber, routers, and network electronics. Biden's 2021 Infrastructure Act (IIJA) also restriected using some certin chinesse technology when buidling internet infrastructure.

[1] https://www.speedtest.net/global-index

[2] https://www.picodi.com/us/bargain-hunting/internet-prices-20...

[3] https://op.europa.eu/en/publication-detail/-/publication/57c...

I'm glad I used litellm in my last project

https://github.com/BerriAI/litellm/

OpenAI and Anthropic were founded by people who sincerely believed in the risk of out-of-control superintelligence

Why would they sell there services to Palantir and/or to the military then?

I think the original looks the best and by a large margin

That is my bad; I was thinking of LastPass[1] where it took them months to fully disclose and explain a very serious breach of data.

1Password seem to have a good transparency track record (I edited the original comment)

[1] https://news.ycombinator.com/item?id=34097142

Bitwarden have in my opinion is one of the BEST business models a user can ask for.

It's open-source, and I can self-host (100% free) and the free version is really, really good too, and then a premium version is $20/year which is very reasonably priced.

Also for cloud hosted password manager, you're always going to have attacks no matter what, but at least they are transparent about it .. (unlike say LastPass, Norton LifeLock, Keeper and possibly others). For self-hosting it might be better security, solely because no one cares to attack it, but it's not going to be more secure form engineering best practices POV (but again I might be wrong .. I'm not a security engineer of any kind)

Not sure why, but they did cooperate with the government on such matters

Facebook previously gave private Messenger chats to Nebraska police, these messages were used as key evidence to charge a mother and daughter over an alleged illegal abortion[1]

[1] https://www.theguardian.com/us-news/2022/aug/10/facebook-use...

Relevant (for some reason though it shouldn’t be; GoDaddy’s track record is that bad.)

Jan 2017: [Godaddy has issued at least 8850 SSL certificates without validating anything](https://news.ycombinator.com/item?id=47911780)

Jan 2019: [GoDaddy injecting JavaScript into websites and how to stop it](https://news.ycombinator.com/item?id=18894792)

Aug 2022: [Tell HN: Godaddy canceled my domain, gave me 2h to respond, then charged €150](https://news.ycombinator.com/item?id=32470017)

Dec 2022: [GoDaddy buying domains when they expire to extort their own users](https://news.ycombinator.com/item?id=34153448)

Jul 2023: [Godaddy just stole my domain](https://news.ycombinator.com/item?id=36854166)

Jan 2024: [Tell HN: GoDaddy Stole My Domain](https://news.ycombinator.com/item?id=39209087)

It's the war on drugs all over again ...

I switched to zero sugar about a year ago, but all the zero sugar sodas use aspartame (yeah yeah not proven to cause cancer, but still not a great sweetener)

for now (out of laziness), I just grab plain sparkling water and add Stur drops

Also didn’t expect to be pulling recipes off GitHub, but I’ll take that any day over those paywalled sites

Definitely want to give this a try!

I don't know if I get you're points because lobster farms are tied to certain external factors in a way things like data centers aren’t.

but either way, the argument feels very NIMBY: it’s not ‘no housing,’ it’s ‘just not here.’

so when someone say ‘let someone else host them,’ it really comes across as: I want the internet, just let other communities pay the environmental cost.

yeah, another way to put it: if you don't want factories, that's fine; just don't buy manufactured stuff .. the same with data centers, if you don't want data centers then don't go on the Internet because by doing so you're becoming part of the problem.

in the last ~2 years:

VSCode/GH copilot -> windsurf -> Zed/Claude code -> Zed/codex -> Zed/opencode -> Antigravity/opencode

I'm only using antigravity cause they have good limits for now .. (but it we be matter of time before it will go away and then go back to Zed)

They’re more active on Twitter/X,

idk what it is about them that every "tech bro" type guy around me follows them, but I never followed them myself, so I was surprised to know they only have 300k on Twitter.

yeah we use an LLM for the grading .. (for the free form questions)

the flow is basically:

When practice questions are generated, the model generates the question + the reference answer together, but the user only sees the question. then on submit, a smaller model grades the learner answer against that reference answer + the grading criteria.

I benchmarked a bunch of judge models for this on a small multi-subject set, and `gpt-oss-20b` ended up being a very solid sweet spot for quality/speed/structured-output reliability. on one of the internal benchmarks it got ~98.3% accuracy over 60 grading cases, with ~1.6s p50 latency, so it feels fast enough to use live.

for math, it’s not just LLM grading though:

- `SymPy` for latex/math expressions, so if the learner writes an equivalent answer in a different form, it still gets marked correct; so `(x+2)(x+3)` and `x^2 + 5x + 6` can both pass. (but might remove that one since it might be easily replaced by an LLM? And it's a niche use that add some maintenance cost)

- tolerance-based checks for the JSXGraph board state stuff; so on the graph if you plotted x = 5.2 instead of 5.3 it will be within the margin of error to pass but will give you a message about it

I also tried embedding/similarity checking early on, but it was noticeably worse on tricky answers, so I didn’t use that as the main path.

It's -kind of- launched, still have couple of things to tight.

And will add a privacy policy by the end of the day, thank you for point that one out

yeah sure, my email is in the bio

https://talimio.com/ Generate fully personalized courses from a prompt. Fully interactive.

New features shipped last month:

- Adaptive practice: LLM generates and grades questions in real-time, then uses Item Response Theory (IRT) to estimate your ability and schedule the optimal next question. Replaces flashcards; especially for math and topics where each question needs to be fresh even when covering the same concept. - Interactive math graphs (JSXGraph) that are gradable - Single-image Docker deployment for easy self-hosting

Open source: https://github.com/SamDc73/Talimio

https://talimio.com/ Generate fully personalized courses from a prompt. Fully interactive.

New features shipped last month:

- Adaptive practice: LLM generates and grades questions in real-time, then uses Item Response Theory (IRT) to estimate your ability and schedule the optimal next question. Replaces flashcards; especially for math and topics where each question needs to be fresh even when covering the same concept. - Interactive math graphs (JSXGraph) that are gradable - Single-image Docker deployment for easy self-hosting

Open source: https://github.com/SamDc73/Talimio

The Stargate money didn’t show up I guess, and now the whole gridlock is collapsing?

I find https://github.com/steipete/gogcli a bit easier (but still confusing to setup)

Google Workspace API(s) keys and Roles was always confusing to me at so many levels .. and they just seem to keeping topping that confusion, no one is addressing the core (honestly not sure if that is even possible at this point)

pre-ai if I had to include Google search queries in a commit, I’d be so embarrassed I’d probably never commit code like ever

Didn't Dario Amodei ask for more government intervention regarding AI?

I highly suspect he might even consider Anthropic since they enforced restrictions at some point on OpenClaw form using there APIs

I switched from using YouTube to invidious mainly because they don't support shorts and blocked YouTube on the DNS level, it's a bit slower, but I know I won't be sucked into doom-scrolling

I mean they are only running a small version of codex can they run the full one? Or the technology isn't there yet?

Yes!

I tried to go as far as 32k on the context window but beyond that it won't be usable on my laptop (Ryzen AI 365, 32gb RAM and 6gb of VRAM)

This is model 12188, which claims to rival SOTA models while not even being in the same league.

In terms of intelligence per compute, it’s probably the best model I can realistically run locally on my laptop for coding. It’s solid for scripting and small projects.

I tried it on a mid-size codebase (~50k LOC), and the context window filled up almost immediately, making it basically unusable unless you’re extremely explicit about which files to touch. I tested it with a 8k context window but will try again with 32k and see if it becomes more practical.

I think the main blocker for using local coding models more is the context window. A lot of work is going into making small models “smarter,” but for agentic coding that only gets you so far. No matter how smart the model is, an agent will blow through the context as soon as it reads a handful of files.

I was waiting for someone to say "this is what happens when you vibe code"

https://samdc73.com/