Just to annoy the non-Australians - Status Quo filmed a studio recording version of Down Down (Prices Are Down), complete with guitars shaped like Coles Down Down signs, and the drummer trying to play while wearing Coles Big Red Hands. It's on the official Status Quo YouTube channel:
I think exactly... and it can be "pervasive": you end up with special-casing None in a lot of places, when actually the "better" thing _could_ be to have a check once somewhere higher up the stack and dealing the None case there. Or (depending on how None is used) restructuring the code so there is never a None case. Often None signifies just "what happened to be coded up first" and that historical legacy can end up enshrined in the code structure for years.
It's not necessarily always "wrong", but you essentially have to think about "what's the right behaviour on None" a lot more, and while yes the type checker will help avoid runtime errors from random places, I think that can lead to essentially runtime bugs if you answer that question wrong. Or, at best you end up with quite messy data transformation and error handling, which itself can end up as a source of bugs.
One thing I try to ask myself is "what would this look like without `| None`", and take it from there. Also, it isn't `| None` as such, it can be any union type: code can end up with blurry responsibility.
> when actually the "better" thing _could_ be to have a check once somewhere higher up the stack and dealing the None case there.
This is what I tend to prefer. Not always possible, but worth striving for.
> One thing I try to ask myself is "what would this look like without `| None`", and take it from there. Also, it isn't `| None` as such, it can be any union type: code can end up with blurry responsibility.
Yeah, unions can be problematic, but at least they're more legible than None as this can basically break everything.
Not pixel art. If the polygons visible when completely zoomed in are any indication, it's 3D rendered. Don't let that detract from how utterly gorgeous and intricately made it is though!
Competing with chinese open-weight model which can run openly or on independent providers. Once the temporary hardware shock is over and becomes more easily available, there will be tons of independent inference providers hosting open-weight models. And I suspect we will see dozens of custom, niche distilations from open-weight models being offered by smaller groups.
Anthropic has no value proposition - just being 3-6months ahead in frontier models vs. open-weight doesn't justify trilion dollar valuation.
People keep saying this about open weight models but I haven’t heard of any real companies any of my friends work at willing to switch over to them.
My company buys Claude because it’s a trusted name with a user-friendly interface and integrations with a lot of business apps.
This “open weight models will take over based on price” is analogous saying that Linux and LibreOfficr will displace Windows and Microsoft Office in the workplace because they’re drastically cheaper.
Another analogy is that people talk a lot about the resurgence of audio CDs, but if you look at sales numbers it’s about 4% of peak audio CD sales from the year 2000. Sometimes in society a trend is more that people keep saying that a trend is happening rather than a trend is really happening.
> This “open weight models will take over based on price” is analogous saying that Linux and LibreOfficr will displace Windows and Microsoft Office in the workplace because they’re drastically cheaper.
You can't just literally drop LibreOffice into an organization and have it automatically and seamlessly work.
LLMs are, definitionally, fungible. Their input is tokens. Their output is tokens. They have minor variations in chat templates and can be trained to perform better in a particular harness (like training GPT in Codex).
Do you or your friends at work use redis? Postgres? Are they hosted at "Redis Inc" and "Postgres Inc"? No, they're commodities hosted at a huge number of places.
Just like you can't just literally drop Qwen into an organization and have it automatically and seamlessly work, not when all your employees are using applications like Claude Desktop and the Claude VSCode extension.
Trust me, I know, I recently switched my own home setup to local AI and it took me a few weeks to figure out an ideal harness setup.
They don't need to "switch" over to them. They just need to stop using frontier models for everything. And I can assure you most big tech companies are already deploying open-weight models and telling their employees to use them more.
Still, Sonnet is way more expensive than an equivalent open weight model. So even if your organization is turning off Fable and Opus they're still paying a premium for Anthropic's software.
I went from using Anthropic via API daily to using Kimi K3 (now 3.1 pro) daily, only going back to Anthropic when a refactor is too much for Kimi to handle (once in a month's worth of work).
That surprised me. I didn't expect K3 to be as good as it was and now with 3.1, it's clear (imo) that the gap is closing. Anthropic still has a minor edge but from the 'feel' of these open models, that gap is negligible, not a moat.
Up in the air at the moment, it's still mostly the developers using the open models, but we'll see.
https://www.jan.ai/ is what I've been using myself outside of work, it's decent, built around "projects" and adding files into it etc., but it's not quite the same thing as what Claude Desktop/ChatGPT Desktop can do. Lightweight and friendly, though.
But for real Claude Cowork/ChatGPT's Desktop Work replacement, if you are okay with Mac-only (my work is) then Goose Desktop is the most complete replacement today.
https://openworklabs.com/ also exists, but I've not dug into it that much personally yet! The space is rapidly changing
The first. There have been massive waves of price cuts in Q3. OpenAI and Anthropic keep slashing prices to take market share from each other, it seems.
I generally don't think LLMs writing prose for humans is a good idea (except for perhaps the most boring kinds of copywriting) but I feel a lot more positive about LLMs as reviewers, even if I don't necessarily use them myself often. My positivity comes from how relatively easy it generally is to validate feedback on a piece of writing you're intimately familiar with.
Gotta say, I don't relate to your concerns about "stray[ing] too far" _in the short term_. I think if reading LLM feedback has upset your original thinking that catastrophically you may have either been wrong (in which case, good that you received feedback early?) or writing with little intention.
However in the long term I can totally see how you could get into a habit of laziness, trusting the LLM to fix your stuff, or mindlessly applying fixes without really considering them.
re: straying too far. I see what you mean but I didn't mean if the feedback itself was to influence your writing
I meant if you were to accept and incorporate the feedback (even if done manually yourself) a handful of times, then at what point have you strayed far enough. Once, twice? If you set it down and revisit, then what if you forget which parts you came up with yourself?
I guess my reasoning here is I never want to guess which is my own ideas/writing. That's too slippery of a slope.
> I guess my reasoning here is I never want to guess which is my own ideas/writing.
Wow! Okay, I wouldn't have guessed that was your concern. How do you feel about feedback from a human editor then? Or is that preferable because it's from a real person? I could empathise with that distinction, TBH.
That is not universally true at all. For example, Milton's Paradise Lost, or much of Shakespeare. There are countless other examples of blank verse in English.
Up until electricity abundance, aluminium was almost unknown to most as refining it requires lots of electricity.
Aluminium ore is very abundant too but we didn't know what to do with it.
Yeah. It's titled "a love song to nicotine" and 80+% of it is about the benefits of nicotine with a claim that "sublingual nicotine’s addictive potential feels low", but TFA links to the tacked-on twitter screenshots at the end. Not the point, but it does make one wonder. Though, to be fair, I only realised that the main body of the article existed when I accidentally scrolled back up "too far". Perhaps the author of TFA received the exact same link and made the same mistake as I initially did.
Dataset of one, but I am struggling to quit sublingual nicotine after using it to quit vaping, which I used to quit smoking. Nicotine in any form is intensely moreish in my view.
It’s not a dataset of one (you’re not alone!), there is a huge corpus of data and nicotine is one of the most addicting substances known to humans.
I cringe reading anyone saying “nicotine is great”—because I’ve been there too and it’s literally one of the most devilish things ever.
It’s absolutely addictive, anyone who says otherwise is suspect
What I take issue with is the original sentence saying it’s “bad for you”, I was expecting new science showing that it’s, well, bad for you rather than “just” addictive
Well, it becomes bad for you if you start needing to take so much of it that the other stimulant effects start causing problems, like poor sleep. Most addictive substances have this potential, since "the dose makes the poison". But the same dose forever rarely satisfies cravings forever.
“Just nicotine”/nicotine salts can elevate blood pressure and heart rate which can be very dangerous for some people and long-term bad if not resolved. But by far tobacco smoking is the most physically dangerous which is obviously not the same as “just nicotine”.
Not studies: Somewhat famously Jessica Simpson got accidentally addicted to nicorette[1]. I heard it as a question in NPR “Wait Wait Don’t Tell Me” (radio-quiz game show) and it ended up being true.
I’ve known many people who used chewing tobacco and were not smokers.
Not making an accusation, but it reminds me of a chat transcript oddity I saw while using opencode back in August. It replace a syllable of a word with a punctuation mark that has the same pronunciation or HTML entity. It was something like "...the &litude of that waveform..." but I can't remember exactly, nor which model I was using. I just noticed it and thought, that's weird. A few days later I read about text watermarking and figured it may be that.
Happens on paid too. Every once in a while, one of my conversation names will have random Chinese characters in them. I guess distillation goes both ways.
Its the letter thorn, which was turned into y in print back in the day because ... cost. It was originally the leading letter in the (or thee) instead of th.
Nowadays, and for quite a lot of the past, people like to riff on "ye olde shoppe" and the like to imply "ye olden dayes".
The problem, as I don't really see it, is that a possible cost cutting measure in printing seems to have swung back and changed the English alphabet. I don't really buy the official explanation for the loss of thorn. I think it simply wandered off and got bored, it looks quite like a p too. I gather that it originally sounded slightly different to "th" but English has largely lost that distinction.
reply