Interesting to note that mimo v2.6 pro has been released at I think 1 trillion parameters and does mostly better or equal than grok 4.7 which is a 2 trillion parameter. Both of which got released on the same day.
That’s a factor of half the parameters. I would be curious to see more on the focus of smaller parameters model and pushing its frontiers
For me it seems like this model’s overthinking should be tamed.
Perhaps instead of giving it a one shot task with vague prompt. I wonder how it can perform with much more detailed and constrained prompt (can you please elaborate more on the level of detailness and ambiguity that the prompt is and where does this model seem to overthink the most?)
Also are there any ways to tame such overthinking of models in general?
I hope that once models start becoming smart enough (I think for me it’s already there) or becoming genuinely the Sota. They then start focusing a lot more on optimizing token usage
Interesting, I would also like to see this curation not just in factor of interests, but also disinterest.
For example, suppose someone doesn't want to read about AI. Instead of blanket banning words or having custom ublock origin scripts (which is what people are doing nowadays), it could ironically use AI (jev) to do that perhaps more robustly.
I think that this is an interesting idea though.
In theory, I would also like to see something which can get the gist of my account (and then find out all the things that I like for example) which could then give the keywords for the curation. This way I wouldn't have to enter the amount of keywords as well.
Really interesting and meta project for hackernews, starred :-D
I started this with the intention to run it as a daemon on my homelab server and send the interesting stories to me on WhatsApp, which is why it is filtered based on interests only.
Oh okay, yea this is a good use case too but for me, viewing hackernews is best from the browser (though I guess sometimes I am reading hackernews too much haha) but I certainly get the appeal of what you are doing. Good luck with the project!
Let me add a few more layers just for some fun and profit :-D
Here[0] is an archive.org page which archives an archive.is page which archives the original article about internet censorship of an archive site that might require an archive of an archive site to read (given that people of spain cant now view archive.is in the first place or might have some difficulties doing so)
> "Temporarily Offline Internet Archive services are temporarily offline. " What did you DO
It seems that all of internet archive was down and I don't think that it was because of me (I hope), maybe just some other issues and some sort of just a chance that internet archive got offline at the exact same time that I had shared this thing here.
but when I had opened up hackernews and I read your comment, I genuinely felt myself also as what did I DO.
It is back again now for what its worth, whew what a relief but folks please remember to donate to internet archive for its posterity :-)
Been using the archive page firefox extension for a year or two. Yet now it looks like a total war against this resource was waged and it's no longer a good resource for archiving.
archive.is is the best substitute. For some reason people here won't accept that. They think the internet is still a calm cooperative place like the olden days.
My office's "net nanny" blocks archive.is as a "Russian site". However, archive.ph (use the same link across all the various sites) is not blocked (yet).
For me, archive.today, archive.is, archive.md, archive.ph, etc. are _not reliable_ for a number of reasons
But some archive.today users who comment on HN cannot seem to accept that archive.today may not work for everybody else
NB. Archive.today is not a "substitute for archive.org". Archive.today does not do www crawls
As for archive.org, I know of a number of alternatives but each is generally less reliable and/or less comprehensive than archive.org
Comman Crawl, i.e., downloads from data.commoncrawl.org, is reasonably reliable but not as comprehensive as archive.org. CC is not a reasonable substitute for archive.org's CDX service. The CC CDX endpoint, index.commoncrawl.org, historically has been easily overwhelmed and unreliable
As for archive.today alternatives (no crawls, only user-submitted URLs), ghostarchive.org seems well-designed but not used much. No CAPTCHA, HTTPS and Javascript are optional and HAR files are provided. Whether it gets blocked like archive.today sites I do not know
NB. Archive.today users may be using archive.today not as an archive but as a lazy man's solution for "paywalls" (Javascript annoyances)
Where that's the case, comparsions to archive.org or other archives that are derived from crawls are inappropriate
Thanks, I will try to look more into this. I think it might be related to number of connections of piping servers but I am a little perplexed about it on how this is possible on an archived link.
Can you please tell me more details so that I can hopefully try to fix it in the future. Did you try to open up the archive.org link of the website or the website itself (which is just a gh page: https://serjaimelannister.github.io/htmlpipe/?https://ppng.i...)
I would like to know more so that I can hopefully fix that for the future, have a nice day :-D
Yes they actually do, but because its cloudflare which is offering this, blacklisting it might lead to blacklisting can be more negative and cloudflare has a much higher incentive to not make these tunnels useless. They are also more powerful and can fix things which would be harder for smaller companies to handle (atleast within the context of cloudflare tunnels)
> The fundamental problem is simpler: no one will slow down because no one trusts anyone else to slow down.
I would argue that nobody trusts anyone else in the case of AI/AI related stuff.
The amount of money involved in the space is astronomical and incentives are really misaligned. AI is such an extremely polarizing topic.
A meta example of this distrust is that I can't (really) trust Dario Amodei's comments itself! (is he saying this for more future IPO money or out of genuine fear) which was ironically also what your comment's about as well.
By following the same chain of events, we can have wildly polarizing claims about the same event and its explainations.
I seriously have to wonder what historians will have to say about this period of human history.
> The reality is, capabilities have largely converged across foundation models over the last 18 months, and much of the value add is coming from the harness layer itself now.
Can you please elaborate on what are your thoughts on open weights models (GLM 5.3, Kimi K3, deepseek etc.)
and if the value add is coming from the harness layer itself, then thoughts on open source harnesses (there are so many harnesses but to name a few: opencode, pi [omp as well], maki, codex is OSS as well, fx.sh) and you can always combine them with skills (Obra/superpowers, matt pocock skills plus using these skills and others to create some other custom skills tailored to your use case as well)
And what about the combination of both now with this cheap open weights models + open source harnesses and other things to compete over the closed garden ecosystems?
How does that comparison follow in reality
Could you in theory use these methods to save on the massively expensive $$$ token spending on Anthropic/OAI?
(Personal anecdote but I have GLM 5.3 + maki [sometimes omp/opencode but mostly maki] and its good enough for most use cases out there that I have and I dont know of too many use cases outside of say recreation of games for examples maybe that I would prefer complete SOTA models. I would also love to know where you believe that SOTA models absolutely do still make the difference discounting the benefits provided by the harness.)
> Look at image gen: you press the magic button and get a 'Pelican on a Bike' - but you can't just change the 'hat' of the Pelican. You have to press the magic button again, and you get a whole different Pelican on a Bike.
I do understand what you mean and perhaps I am trying to treat it as a problem to be solved and challenge accepted but my first thoughts are if its an vector image like SVG (the famous simonw's pelican on a bike benchmark)
Then you could in theory have a layered approach and then just change the code of the hat so for example (IIRC) <--Hat--> Code. <-- Bicycle-->
So basically I am suggesting modularizing of concerns of areas so that you could have better autonomy over what exact thing you wish to change.
Though I imagine that you and I might be saying the same thing and you are suggesting that the crux of the argument is exactly that you need specific know how in reaching to that said modularity where you can best use AI.
When you might need the generation of the SVG as compared to image generation because you realize that your project might need the pelican changing lots of hats and image gen wouldn't be feasible but for that to actually say to AI, You might need to know in the first place that SVG or (HTML?) might be better use cases for this. Again, I must admit that I am not an expert in this so I can't absolutely comment on what the best thing for this particular use case could be and maybe that might be your point.
Have we arrived at similar conclusion or perhaps, is there more nuance to it?
On a side note: I actually once had a real use case of where I needed layered approach similar to Figma but generated through AI. preferably something which can just work through good ol chat app UI which could give an index.html or other code for this modification/layered approach.
Like recently there is https://bento.page which has been somewhat similar for this in some sense but for pdf's. So is there something different but for image-ish thing? If there is an expert lurking here, I would love to know the answer and gain some knowledge about it, thanks :-D
Yes, for things we can construct, and where we can train on the pieces, there's hope - software is a bit like that. Its not like 'raw' image generation, it's not a perfect example but the notion remains.
That’s a factor of half the parameters. I would be curious to see more on the focus of smaller parameters model and pushing its frontiers
reply