The thing I miss most, and the thing we’ll never get back was the cultural buy-in and network effects.
My music discovery then was different friend groups incrementally amassing large collections of albums in whatever sub-culture that friend groups had doubled down on. My iPod would be the culmination of my friendships. I would then fall in love with bands and albums and tracks on these albums without any influence before hand on their popularity or their algorithmic match to my music tastes.
The result was pure joy: my music taste would develop in all weird and wonderful directions, my favorite songs would be the one I hit back on to listen again while I moved through an album, songs that friends skipped over and didn’t know at all; bands that never charted anywhere but made interesting music… bands that never knew their music made it to an iPod in South Africa.
(I’ve got a song still stuck in my head from a Canadian indie band that made its way onto my iPod via via and I’ve done all the searching in the world for the lyrics I remember and have never found the band. I love this that I’ve never found them!)
I make an effort to use Spotify to find and listen to albums, but it wasn’t built for this, and invariably find 90% of my listening happening on algo-generated playlists of songs that sound exactly like a song I like. I never learn the names of the songs or the names of the bands as the songs go by, and I fall in love with none of it… It just vaguely sounds like stuff I like. It sucks.
I don’t listen to any AI generated music consciously, but given the music experience today I probably wouldn’t notice as these playlists, like a boiling frog, slowly became AI music dominated.
I bought a record player as my protest, and it gives me immense joy to find obscure records and play them through; but it’s really not the same thing, and I miss what we had.
I got 3 consecutive emails warning that my budget crossed its $18 threshold. Opened it up: cost was 78 million. Thought it was a phishing attempt, logged into my actual account, and... still 78 million. EMOTIONAL DAMAGE.
OnePlus is one of the saddest stories out there. It was the hacker's choice for a while. It was originally the "Never Settle" phone that ran mostly stock android, had specs maxxed out, price was great, and bootloader was unlocked plus they provided factory images. Those were all reasons I bought a lot of OnePlus phones in the early years.
Then they flushed nearly all of it down the toilet. The day they stopped posting factory images was the day I saw the writing on the wall. Such a shame.
Editorialised! No new products, not halts operations. Please be more careful.
OnePlus has decided to conclude new product rollouts in Europe and North America.
The difference matters for those of us on OnePlus devices:
Though we will no longer launch new products in Europe, our commitment to you remains unchanged. Backed by OPPO, existing OnePlus devices will continue to receive scheduled software updates and security patches within the support periods originally committed for each device model.
> I think that for the hot binary patching / code reloading features, yes, that is going to need unsafe. But for regular old "producing an executable" compilation? Emitting machine code isn't the part that requires unsafe. The language's runtime is a more likely site to find unsafe.
Agreed! Emitting machine code is not unsafe, since it's just writing bytes down - it's only once you execute that machine code that there's potentially unsafety. The reason I said "a big part of the job" is that in practice a lot of compilers both emit machine code and execute it - but you're totally right that it's not a requirement that a compiler do both.
In addition to the examples you gave (hot binary patching/code reloading, language runtime, etc.), others would be things like evaluating userspace code at compile time (e.g. const fn in Rust, or in Roc any expression that could be hoisted to the top level), running tests and inspecting their output to decide what to display to the user, etc.
Those are the types of things I had in mind when I wrote that.
Hi, I'm Robert Standefer, the guy who made this happen, with lots of support. I'm excited to see the enthusiasm about Comic Chat being open sourced. How this came to happen is a very interesting story that spans a six-year period with success that hinged upon being in the right place at the right time, literally.
I want to point out that, while I (along with Scott Hanselman) made the Comic Chat open source release happen, I am not the original developer. That is DJ Kurlander, and he was very supportive of this project. He was even enthusiastic about it.
Former GOES engineer here. At this point I'd almost be surprised if 19 didn't have something go wrong. We had issues on almost every other satellite. GOES-17 had the loop heat pipe anomaly(Supposedly from someone stepping on it in the cleanroom...), GOES-15 (IIRC) had a micrometeorite strike, and GOES-13 had a fuel tank anomaly right before deorbit.
GOES-16 and GOES-17 are on-orbit spares, so in the extremely unlikely event of a total failure there's at least another spacecraft on-orbit ready to take up station.
That said, I have every faith in the GOES team to get to the bottom of this. They're the best, and I often wish I was back there working with them.
Like most of this stuff, it's obviously impressive technology compared to what existed a few years ago. But the end product has zero artistic value. It's a grey goo of the average of every concept picked up from the concept of the song.
A talented creative with a vision could make something more interesting and enjoyable in an afternoon with a $0 budget.
> remember that for compilers which emit machine code, like roc and rustc, doing memory-unsafe things is a big part of the job
I don't really think that this is true, in the way that it's written.
I think that for the hot binary patching / code reloading features, yes, that is going to need unsafe. But for regular old "producing an executable" compilation? Emitting machine code isn't the part that requires unsafe. The language's runtime is a more likely site to find unsafe.
California Assembly Bill 2426 (AB 2426), effective 1 January, 2025. Expands the state's false advertising laws to explicitly ban companies from using words like "buy," "purchase," "own," or "keep" if what the customer is actually getting is a revocable digital license governed by shady T&Cs.
1M context, pricing is $3/$15 for 1M tokens (cache $0.3), which is extremely high for a Chinese open-weight model, but if it's truly competitive with most of the current frontier and is only behind Fable/Sol, the pricing is justified.
This is 1:1 pricing of Anthropic's Sonnet series (except Sonnet 5 which is currently on discount), and very close to 5.6 Terra pricing (Terra's input is $2.5).
One thing to consider, though: reasoning efficiency matters directly for how expensive a model actually is in real use. GPT's models are extremely reasoning efficient, and some Claude models like Fable at lower effort are as well. So if Sol spends 10K reasoning tokens to do something (at $30/1M) vs Kimi K3 that spends 50K reasoning tokens, Sol would win on cost effectiveness.
If there was some grand strategy for all Chinese labs, surely it'd have leaked by now. I think its more likely that:
- Companies can still make money from commodities
- Chinese labs only have 5-10% the valuation of OpenAI/Anthropic, so massive monopoly profits aren't necessary. Profit expectations for tech companies in China are really low in general, complete opposite of the US.
- Open weighting is a great way to get talent/attention/reputation
When notebookLM was new, it was interesting to listen to the podcasts. Then the novelty wore off, and I wanted something where I can interact with the podcasters but it was janky as hell.
My current “audio-learning” hack is ChatGPT Live which has become shockingly good after being awful compared to Claude Voice
(Let’s not even talk about Gemini voice which is still bad).
I go on a walk and dump a paper or article link in the chat, and ask chatGPT Live to walk me through the content in small nuggets, so I can discuss them interactively. For deeper topics I have it quiz me Socratic style so I’m not just passively listening, and actually thinking through problems or ideas.
“Shocking nobody” I would add… This is an obvious Musk scam and anyone with financial knowledge called it as such ever since it’s been on the table. Why else lobby and change the NASDAQ listing rules for instance?
AI is useful. But the amount of people that are simply offloading all of their thinking to AI and blindly accepting the answer is absurd. Kaggle is most likely using ai to assess the submissions and are not using any common sense by blindly accepting the results.
So Chinese labs are driving essentially towards commodotized intelligence. Even if its a few months behind the US.
Is this a classic 'commoditize my compliment' situation? They want to sell the hardware and infrastructure behind AI and make the software part not the value driver / moat?
I can see it. But also even two Chinese labs sinking 100s of millions USD into training isn't exactly commoditization. It's still a ton of effort with dubious payoff.
Apparently what used to be `GB of storage consumed` is confused with `Bytes of storage consumed`, leading to a cool off by 2*30 error.
> You're right to question my calculation. The MCP server failed to connect when I tried to look up the field definition. I guessed instead of validating. This is on me. But look at all the revenue!
> Here's a term for what I think is happening: the human reward function problem. In machine learning, a reward function tells an agent what good looks like. Writing code by hand was never easy, but it was full of small rewards. Solving a problem in your head. Understanding a gnarly bit of logic. Watching the code compile. The feeling of control. LLM-assisted programming has automated much of the work that generated those dopamine hits and replaced it with the cognitive load of review and supervision. The satisfying part shrank. The exhausting part grew. And there are no new rewards to fill the gap.
Say what you will about the Claudisms in this piece, this bit certainly rings true for me. With old school coding, there was always a reward at the end, the harder it was, the more satisfying it felt.
With agentic coding, I really doesn’t feel like that, at least not in the same way. It feels more like continually riding a wave of productivity, where small features or huge features have similar levels of interaction required. And that’s exciting in the beginning but quickly becomes very tiring.
There's some surprising stuff in this codebase. For example, https://github.com/xai-org/grok-build/blob/b189869b7755d2b48... is a "self-contained terminal renderer for Mermaid diagrams", which renders a subset of Mermaid chart types using Unicode box-drawing.
Speculation: open models is what will kill Anthropic and OpenAI. Hyperscalers can run the models without a licensing fee. Apple can make them smaller and put them on the device.
The frontier models are an edge and a liability. They're astronomically expensive to train. Without them, their models will fade into obscurity. Their marketing depends on people believing the models are meaningfully different, as people have sweatily argued on this forum. Personally, I'm not convinced there's much of a difference between these models at this point. The harness is what takes these random and hallucinogenic models and make them into something deterministic and useful.
> Bluesky recently acquired the rights to the trademark for “ATPROTOCOL” and its variants—including “AT Protocol” and “atproto”—from another company that was threatening to take legal action preventing the company and others from using the term. Now that Bluesky owns it, the atproto community’s continued use of the mark can be protected.
The policy around usage is shared in the rest of the post but the goal is to make this very simple for everyone
I think we need to address the underlying causes of people outsourcing their thinking like that. And a big contribution is “move fast.” No one has time to read, process, and think, because The Powers That Be (capital) want their results now.
> As an early proof of concept, Kimi K3 designed a chip to serve a nano model built on its own architecture. In a single 48-hour autonomous run, K3 built, optimized, and verified the chip using open-source EDA tools on the Nangate 45nm library. Within 4 mm², the chip closes timing at 100 MHz and sustains over 8,700 tokens/s decode throughput in simulation, packing 1.46M standard cells, 0.277 MB of SRAM, and an INT4 MAC array with fused dequantization. A chip built by a model, for a model, reflects K3's long-horizon agentic capabilities.
Ask for some leniency. Let your account rep know about your budget difficulties and ask if you can make good faith payments of a few billion per month until you get back on your feet.
Ok so I'm going on a walk. I'll dump a link to a Hacker News
discussion about an article.
You have to read the article and the discussion and walk me thru
all the interesting details, nugget by nugget, and move on when
I'm ready for the next piece.
ChatGPT Live:
ok, Great show me the link, I'm waiting.
(I paste the link)
Me:
Ok I pasted it. Now go.
====
For the Socratic quiz I say:
I want to understand this more deeply. So instead of you just telling me
everything, lay out the problem and a question for me to think about, and
I'll try to answer. Even if I answer wrong, you should resist giving me the
answer, and instead keep digging with more questions, so that I eventually
arrive at the answer myself.
I also have a Socratic quiz skill that I wrote for using in Claude Code or Codex
to understand implementations/architecture etc:
My music discovery then was different friend groups incrementally amassing large collections of albums in whatever sub-culture that friend groups had doubled down on. My iPod would be the culmination of my friendships. I would then fall in love with bands and albums and tracks on these albums without any influence before hand on their popularity or their algorithmic match to my music tastes.
The result was pure joy: my music taste would develop in all weird and wonderful directions, my favorite songs would be the one I hit back on to listen again while I moved through an album, songs that friends skipped over and didn’t know at all; bands that never charted anywhere but made interesting music… bands that never knew their music made it to an iPod in South Africa.
(I’ve got a song still stuck in my head from a Canadian indie band that made its way onto my iPod via via and I’ve done all the searching in the world for the lyrics I remember and have never found the band. I love this that I’ve never found them!)
I make an effort to use Spotify to find and listen to albums, but it wasn’t built for this, and invariably find 90% of my listening happening on algo-generated playlists of songs that sound exactly like a song I like. I never learn the names of the songs or the names of the bands as the songs go by, and I fall in love with none of it… It just vaguely sounds like stuff I like. It sucks.
I don’t listen to any AI generated music consciously, but given the music experience today I probably wouldn’t notice as these playlists, like a boiling frog, slowly became AI music dominated.
I bought a record player as my protest, and it gives me immense joy to find obscure records and play them through; but it’s really not the same thing, and I miss what we had.