Hacker Newsnew | past | comments | ask | show | jobs | submit | nl's commentslogin

I believe most speculation is that Opus is 5T model, Fable a 10T

Elon claimed Opus was 5T in April, and I think it's fairly likely this is accurate: https://x.com/elonmusk/status/2042123561666855235


This isn't necessarily true.

While every formal proof can be encoded mathematically that doesn't imply that the grammar of lean or rocq is sufficient to encode every potential proof.

However the core of what you are saying: that every possible proof exists in the space of all mathematical statements is correct.


> The MySQL purchase. WTF, this was horrible and stupid. The only interesting effect of this was to make Sun a target of acquisition for Oracle.

Strong disagree with this!

The MySQL purchase wasn't a lot of money and was a move towards becoming a systems company (like you suggested they need to be)

Oracle bought Sun for Java, not for MySQL.


My memory is that Oracle bought Sun for the Java lawsuit, not Java itself.

Oracle sued Google over Java - it wasn't some previously existing case.

Yes, it is Oracle so there's a great chance they were considering this when they bought it.


The disagreement had been going on for years (at least since Google launched android, perhaps before that), Google and Sun had been in “negotiations” that were publicly documented as going nowhere. All that had happened is Google implemented more “workarounds” to use less of Sun’s IP.

I actually remember media speculating that Sun didn’t want to take the next step and actually file a lawsuit, because they didn’t have enough resources to go up against Google in court.

Then Oracle bought Sun, and suddenly the unproductive negotiations transformed into a lawsuit.


And thanks to that lawsuit, we got to see Oracle lose (always nice!) and SCOTUS establish that using APIs constitutes fair use (awesome!)

My problem with this is that yes, looking at the species level from our sample of ~1 the most intelligent species wins out (although there seems to be a lot of debate as to if Neanderthals were more or less intelligent than humans).

But within the sample of the human species it is more common for less intelligent people to control more intelligent people.

This happens within political systems (political leaders are usually above average intelligence, but hardly the most intelligent) and within companies and other economic systems.

I don't see this fairly obvious point discussed, and not really sure what it shows.


> not really sure what it shows

it shows that you don't get anything for free. If it was without any cost in other areas to make a human massively more intelligent, it would presumably happen over time via the same processes that led to us.

As it is, the higher (and lower!) ends of the distribution tend to be highly correlated with other issues (mental+physical), and the further you go, the more unfortunate things which make it harder to do stuff in general start to crop up like psychosis, autism, ocd, anxiety, addiction etc


I'm not sure.

Maybe there are other things ("leadership", "likeability", "appearance" etc) that are more important in the pursuit of power than intelligence.

I think it's telling that it is usually conventionally intelligent people making the claim that intelligence is what leads to power.


Sure, but hiding the contents provides a lot more value than hiding some parts of metadata.

It's ridiculous to argue otherwise.

It's true that general resources could be cached in some cases, but personalized information couldn't be (and yes was served over HTTP, not just HTTPS).

> It really doesn't matter what's inside

It certainly does matter what is inside, and it's trivial to think of cases where it does. Hiding that content is much much more valuable.

> We kill people based on metadata.

Sure, but they also kill people based on content.


> It's ridiculous to argue otherwise.

I wholeheartedly disagree :)

If you will refer to the “survivability onion”: https://en.wikipedia.org/wiki/Survivability#/media/File:Surv...

                 Don't be there
     I am here → Don't be seen
                 Don't be acquired
  You are here → Don't be hit
                 Don't be penetrated
                 Don't be killed

One will note that it's called ‘HTTPS’ and not ‘HTTPP’ lol

> If you will refer to the “survivability onion”

> I am here → Don't be seen

My point is that you aren't. For example under the old HTTP only internet they could see you were visiting HN, who you were and what you published.

Under HTTPS they see you are visting HN.

It's strictly better except in cases of visiting highly popular data which was cached in public caches. These public caches were much less wide spread than you appear to believe (for example in Australia the Internode ISP tried one for a little while but abandoned it because of the problems with it serving info one logged in person to another - cache invalidation was frequently buggy)

> One will note that it's called ‘HTTPS’ and not ‘HTTPP’

Not sure who that is aimed at but I didn't write HTTPP, and yours is the only mention of it on this page.

> lol

If you are going to lol you had better make sure you are correct about what you are laughing about....


> These public caches were much less wide spread than you appear to believe

You can keep your belief or disbelief. I ran my own, transparently with Squid on my FreeBSD PF gateway box back in The Day when we had a cable ISP with a restrictive data cap. I could send my roommate a link, or reload a link myself any number of times, and it never left the edge of my network after the first. It was really cool.

These days you don't even need to actively do anything for most operating systems' default configurations to loudly announce your presence and behavior to the ‘Room 641A’s of the world:

- Join any Wi-Fi network and your OS or browser (or both!) will automatically hit a well-known captive portal test URI.

- Most OSes will hit an NTP server that reveals the OS maker (depending on if it's Google's or Apple's or Microsoft's or the NTP-dot-org pool) as well as providing a heartbeat to tell the network you're still physically present when you make subsequent requests every ${default-periodicity} minutes following the first time sync following the captive portal check.

- Then once TLS' regressive dependency on synchronized time is satisfied, many default OS installs will hit a weather API whose transport layer is “secure” but whose response size is, if not unique among all possible geolocations that could be requested at the same timestamp, at least a piece of identifying information to combine with others. If I were NSA I would be making requests to APIs like Foreca for every possible geolocation at every ${change-frequency} seconds and storing every bit of sidechannel metadata about them forever for comparison with other recorded traffic.

- And most people's machines will announce their unique collection of software to the world through the act of making network connections to the known networks for the background update checks and “““telemetry””” (remember when the term was “spyware” for anything that remotely reported user behavior?) that have sadly become normalized in modern software.


I too ran Squid. It was more annoying than not (except for images, where I do think it was quite useful)

It seem you concede the point about survivability though?

Those other things that seem unrelated to the utility of HTTP vs HTTPS.


if you visit wikipedia.org/whistleblowing, what exactly value is there from hiding content but force revealing you visited it, making sure it can't be retrieved from a cache within your organization/school network?

and doesn't pervasive tls termination mean that Cloudflare/etc sees plaintext anyway? do we really believe that palantir isn't tapping into that?

i can imagine just a few scenarios where hiding content on the web actually achieves something. mainly stuff like webmail, web chats, banking.


> if you visit wikipedia.org/whistleblowing

Just visiting, sure but if you edit it to leak a secret the protection is everything.

Say you are on your school network and leak something damaging about the school on Wikipedia.

Prior to Wikpedia switching to HTTPS, both the school and any intermediate network could see

  - you visited Wikipedia
  - you edited Wikipedia
  - what you wrote on Wikipedia
After HTTPS, your school could see

  - you visited Wikipedia
That's strictly better in every way.

> if you edit it to leak a secret the protection is everything.

It’s really not, because knowing when you edited it is already enough, given that there’s a timestamped edit history. And the size and shape of your request payloads will show that you edited it, not merely performed GET requests.


Doesn't encrypted DNS make a lot of that metadata moot ?

And of course the additional options available


No, it doesn't. You still see the domain name in the TLS headers.

Encrypted Hello will close that bit given time.

He fled the US and got stuck in Russia instead of Ecuador.

I'm not sure that materially changes the point does it? (Unless the implication is that by saying "he fled to Russia" it somehow implies that he was on the side of Russia instead of "anywhere he can be safe from the US").


The OP was right - avoiding a unfair trial is still fleeing.

The word "flee" is not a pejorative in any way. For example flee North Korea all the time.


No. It is pejorative in the way it is used. Context is important.

If you aren't already familiar with it, the great vowel shift in English is also very informative about the discrepancy between pronunciation and spelling: https://en.wikipedia.org/wiki/Great_Vowel_Shift

The TL;DR is that vowel pronunciation was changing in English between 1400 and 1700, and English spelling was becoming standardised between during 15th and 16th centuries. Depending on if the spelling of a particular word was standardised before or after the vowel shift means the spelling might or might not reflect the pronunciation.


You really, really need different skills depending on the model.

If you are using Qwen 27B you need very prescriptive skills.

If you are using Astra you usually want very minimal skills (because it will follow them but be unnecessarily constrained in some contexts)

If you are using Fable then it depends - it will take the skills as general guidelines but ignore them a lot more than Astra does. Sometimes this is good, sometimes not at all.


Right. That's what modular guidance documentation is for. You could have astra.code or qwen.code.api. All reusable in different workflows. Prescription doesn't belong in the skill itself.

This sort of works.

The problem is that you'll end up wanting to run 2 or 3 (or 20, 100, 10,000) agents at once and that gets very hard with a single VM.

There's also an argument that you should be using a separate sandbox for each code operation a LLM performs (or at least each set of related operations). That's even harder to do with conventional VMs.


I just use git worktrees in a single container. Albeit I don’t have more than 3 agents going at a time generally due to my own context switching limitations.

It's a different level of isolation, worktrees help agent work on different code repository in parallel but things get wonky once you consider processes and environments variables

The way I do it is each worktree gets its own .env, with dev credentials to whatever it needs access to, and my coworkers have theirs set up to have an isolated Postgres database per worktree, but I didn’t bother with that in my setup— maybe I will, though. It’s a simple script to create a worktree and properly prepare it.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: