benjiro29

People keep forgetting that over the last 6+ months a lot of increased action has been taken by OpenAI and Anthropic to detect and combat distillation. Several are public known.

Combined with how short of a time Fable was around before K3 got released. I do not see how the data Moonshot is supposed to extract in such a short notice, that will enhance the model to such a point.

It sounds to me a lot of cope from the US, so they can give this as a reason to ban Kimi models from the market.

OpenAI/Anthropic their advantages used to be:

* Early growth advantage

* Access to a lot of client data to train upon

* Access to a lot of hardware to train upon

Several of those advantages have been eroded over time. That barrier has been shrinking. The US is not the only spot with a bunch of smart people (ironical seeing how many Chinese work in US R&D).

Thing is, even IF they distilled from Fable and got the model so trained up, it means that K3 is a base for future model development. The cat is already out of the bag with how good the model is. When the model gets released on the 27'th, any Chinese company will be able to train their models against K3 openly.

We are not in the past anymore, where DeepSeek was a unexpected hit, but where the Frontier models their advantages (compute, data, growth) prevented more Chinese models from growing.

As somebody who did plenty of scraping for his own little projects, lets just say that reddit their security concerns are just PR for "we do not want to keep supporting old.reddit".

While yes, you can not simply scrap new reddit as easily as a pure html scraper is cheaper to run. And while the new reddit its js "slow down" scraping as it need to run over a headless browser. Do you slow scraping down a lot?

No. Because you can simply spin up more instances and route over more proxies. Given the fact that you can scrap millions of pages per day, on a single cheap 1 CPU node...

The limiting factor for scraping a website is not html vs js / client vs headless browser. Its avoiding detection by running from different ips, faking browser information and masking your traffic to not give off the smell of automatization (and avoiding tls fingerprinting).

I wrote this type of stuff before we had AI to help write it, in a few days time. Now with a LLM at your fingertips in a few hours, you will have a complete working version with cloaking build in.

And the larger size difference does not matter, you can block irrelevant files via faked cache responds, to give the server the impression that your a repeat client, and no need to flood your browser instance with images, fonts etc. Some think that this is a defense, aka throttling you with increased bandwidth usage lol

So in my personal opinion, its not about dealing with scrapers, but trying to push people to new reddit, so they can finally phase out old reddit.

PS: So much fun reading new reddit on your phone browser, to have it pop-up a "use our app" banner, that prevents you from interacting with the website anymore. And even blocks firefox its scroll ability (and thus accessing to the browser its nav bar). Tip for people: delete the sites cookies and cache regularly, to prevent this usage tracking, so the banner newer shows.

!! Be careful when testing the model.

A lot of people are testing it, and reporting disappointed results / benchmaxxxing claim. But do not realize that thinking has a issue with the default configuration.

Important - make sure that THINKING is enabled. By default it wasn't although I was passing the flag --default-chat-template-kwargs '{"enable_thinking": true}' in vllm recipe. The generation_config.json file that is included has by default max_new_tokens as 32k which seems to be cutting off thinking altogether so increase it. At first I was very disappointed with the output I was seeing, but once thinking is enabled, the code quality seems to be MUCH better. More real world testing to be done.

https://www.reddit.com/r/LocalLLaMA/comments/1v2pg99/laguna_...

The interesting aspect of this attack and the how guardrails prevented them from analyzing the attack.

When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker. We ran the forensic analysis instead on GLM 5.2, an open-weight model, on our own infrastructure. This had a second benefit: no attacker data, and none of the credentials it referenced, left our environment.

While falling back to their local hosted Open Weight model, gave them the ability to analyze the attack, while also keeping their internal structure / data private.

The problem with these type of new / transcompile languages is not the langue, its the lacking libraries and 3th party assets. Great if just want the most basic hello world programs, but the moment you need ... for example a http server. Then your often with:

a) none

b) http 1.1

c) ... forget about more

Maybe you need rar support ... O well, ... And then you always enter the world where you need to start linking external libs, and then your mixing not just the base language and the transcompile but also whatever interface for calling those external libs.

Ironically, with LLMs being around, your often better to just have it write whatever is missing. Until you find out then, that the language misses some feature.

O great, you needed crypto support. Ok let the LLM write it for you, its only 100x slower then whatever was written years ago and had tons of enhancements and edge cases fixed.

At that point, you can just tell a LLM to write in C from the start, and be done with it. lol.

Transcompile languages have always had tons of issues, and there is a reason why non became popular. Haxe comes to mind. How many years this exist and barely anybody knows it. Because your often just better writing in the original language and fixing the issues, or picking a better fitting language for your project.

In regards to your post and the 16k reasoning output.

Try setting reasoning levels yourself manually. We see in the benchmarks that one of the graphs shows low, mid, max, so its clearly there.

I had the same issue with GLM 5.2 only offering high/max.

By playing around with openai compatible protocol, and setting the reasoning level from none, low ... high, xhigh and testing a flawed logic test.

It was easy to see that GLM had all the different reasoning levels. Low was like one line, medium did a few, high started to really expand, xhigh was a page or 2, max was MAX.

Very sure that you can force K3 into using less reasoning.

Open Weight release is on 27 july.

Full benchmarks in Mandarin:

https://mp.weixin.qq.com/s/V4xhEIy8xDXSMDPrPkmUAQ

Translation:

https://mp-weixin-qq-com.translate.goog/s/V4xhEIy8xDXSMDPrPk...

Cheaper then GPT 5.6 Sol (according to their results) ...

Where did you hear about the deepseek release?

* Tons of gray testing going on for the last 2+ weeks (people at random getting the new v4 model for a while before its removed again).

* It also DeepSeek their 3th birthday this Friday.

* The its been almost 3 months from the v4 DeepSeek release, and the model everybody have been using, was not post-trained. That is what they have been doing during this time.

People trying out the new DSv4 via the web chat with quick game creation tests. People pulling out stuff like Stellaris clones etc.

https://cct124.github.io/HORIZON6_DEMO/

https://www.showyourcode.app/zh/share/pmpwkamrnai2ue

The Battlefront like game is impressive. Sure, the soldiers are backwards and the graphics are still kind of basic. But the entire movement system (run/walk/crouch/jump), gun mechanics, grenades, capture points, AI fighting / capturing back, etc ... Ended up playing it way too darn long lol The text is in mandarin but its not too hard to figure out the menu. Sniper is OP ;)

The Horizon 6 game has everywhere mesh colliders, shows when you off track dirt being kicked up, etc ... In general, both example are very well polished minus the reverse soldiers issue.

And the price is supposed to stay the same (beyond the doubling during Chinese workhours), because everybody got that update.

Models are not like a fps counter where losing a few percentage has no actual impact. These percentages may be the difference between a model that writes code and one that goes into loops, and think rm -rf . is a good idea.

There is a reason why most models try to stay in the FP4 or higher range, because the reduced accuracy can have major consequences.

You are better off with a 8b FP4+ model then a 27b Q2 model.

Resell value takes a nosedive on modified macbooks...

There is a large culture in the Mac community that anything that modifies your device, may have a impact on your expensive warranty when you need something fixed. Apple has taught their customer base well.

This translates into Macbooks that have been modified often having less resell value. See how many time people with the thermal mods need to point out warranty, battery, and the ability to remove it.

Laptops in general, be it macbooks, windows device, ... are already difficult devices to sell (in my experience) because they are very personal. So anything that steps outside that norm, makes it even harder to sell.

If your planning on keeping your device until eol, modify away. But a surprising large amount of people buy these devices with the idea of reselling after a year or two (as they hold value better then win devices), so they can then use that money on a upgraded version.

If your in EU, you contact the local EU consumer group where you buy the device.

https://www.europe-consommateurs.eu/en/who-we-are/about-us/e...

And file a complaint... As that breaches a dozen or more EU laws. If a lot of people do it in all the countries, it becomes a national issue.

That is the only way you fix things, and yes, we have had multiple successes with companies taking the piss. Even Samsung can not escape as their have officies in the EU and sell products there.

For the folks outside the EU, ... Its a harder fight and you need to look up your local agencies.

The problem that often the victims also have not educated, have no worldly experience, often have no idea about money handeling beyond small items.

This can result in them being exploited again by even more unscrupulous people. The articles clearly mentioned how difficult these cases are to deal with. While they do not go into detail, the above is why.

Its very easy to gain peoples trust when they have no sense of normal anymore, and can you sign this paper, o, we need to go to a friendly notary to help with it. and before you know it, the people just handed over their apartment / or whatever.

There are a lot of good people with will want to help but it only takes one rotten apple to destroy peoples live again. Recently in Europe there was a case of a helper that took elderly their IDs and helped herself to their money. She made 100s of victims. Now imaging that type of person with somebody who probably did not have any proper education and normal independent life experiences that we all had the luxury of having.

In a ideal world, we have proper state funded solutions, with proper oversight to help people integrate into society. Reality is that if any services exist, they are underfunded, often lacking oversight and people fall into the often chasm of cracks.

These type of stories are never clean white and black, but a mix of gray sludge, where we all hope for the perfect ideal solution but often there are not many options. And naivety tend to often do more harm then good.

those satellite lines represent humans being able to connect with modern technology out in the Sahara or in the ocean

Or basic locations in Europe, that can be as close to 20km from a big city. There are a ton of spots where you have at best spotty 4G coverage or no coverage at all.

Used to live where we had 1Mbit ADSL, and no cell ... Trust me, you feel the limitations of that. Keeping your PC running 24/7 to download/buffer, so you can use your day traffic for more important stuff.

Starlink is a insane deal in my eyes. Sure, it uses a lot of power but your paying the same as typical vDSL in Germany. And ironically, ... Starlink is faster then the fixed lines we have here in the "3th" biggest city in Germany.

People really underestimate how much underinvestment there has been in Internet connectivity even in rich countries.

I can’t get Claude to do anything even a fraction of this complexity.

I remember writing a postgresql compatible DB with Opus 4.5, that used S3 as storage and local caching to make it speedy.

Ironically, Opus 4.5 is by todays standards is antique. If you have some knowledge about Databases, it goes a long way.

But you need to do it step by step. Getting the core to work, getting a Pratt parsing going. The whole pgwire protocol ... the data format ...

Step by step ...

With todays Models, your can probably get away with using /goal and telling it to make a postgresql compatible database, while having it run a few days.

Now, making a fun database test project and having it production ready! Big difference!

If you're acting in good faith and your accountant does something crazy or evil, your liability is limited to some extent.

From my understanding, you are the person signing off on the paperwork that is submitted to the IRS.

There is this cache 22 with taxes.

You are responsible, but you outsource it to a accountant. Because you are not knowledgeable about the taxes. But you are expected to be knowledgeable to understand the tax documents that you submit to the IRS. That is why the accountant always ask you to review the documents and sign them like 20 times.

The same applies when you run a company, depending on the country, you need to prove yourself knowledgeable in accounting, before you are allowed to run a company. Normally that is included in a university degree, but if you have a middle school diploma, you need to do a official examen to get that degree.

Whatever you submit for your company, you are again responsible. Even if you hired a accountant.

So while technically, if a accountant makes gross mistakes, the bill will always fall in your lap, because you are expected to understand the reports you submit to the IRS. And catch any errors before doing so.

With the IRS, the burden of proving your innocents is often put you. Its because the good faith argument can be misused easily. That is why the buck stops at you.

So using a LLM or a accountant, really does not matter. Sure, a accountant can go to jail if there has been major issues (its not going to be with one client issue).

But you can lose your house / company, have your life ruined by whatever you submitted.

I have a hard time imaging how those people can drive with the eye tracking features. Beep beep beep beep ...

Do people like zone it out like parents with kids are able to zone out their little screaming devils? lol

on a curve

O yea, that is driver lane assist ... A Toyota rental had the same issue. In a specific steep exit corner (that goes up facing the sun), how many ** times the lane assist tries to force the car to go straight (as in, off the hill! ). The first few times when it happens, scares the ** out of me.

Another fun one is going down a hill in a Rental Opel, roundabout with some cars, no problem. Slowing down naturally, while i see the cars accelerate to enter the roundabout. No need to break as by the time i get close, the cars will have started to accelerate. So my speed will have matched the last vehicles speed by the time i am close. Suddenly, emergency break slam on !!! Because "the car was going to hit the cars in front". Like, wtf!! That created a extreme dangerous situation if there was a car behind.

I really see no benefits for a lot of those new safety features. The old ones like traction controle etc, great, keep them. But all this external monitoring, internal monitoring ... If your a safe driver, those features can make it more dangerous.

If you measure "function" by the relative economic and military power of the country, then the EU has overwhelmingly degraded the function of its initial members compared to when they joined.

Very sure that when the EU was still in its infancy, we had only "west Europe" in arms, vs a USSR (aka all the eastern states and Russia). Now all those states are part of NATO and the EU.

Instead of the border to the closest hostile nation (Russia) being barely 100km from here, its now over 1200km to the first contact point.

That same Russia can barely deal with a Ukraine, that has some spare change backing from the EU. How is again at a war economy? Ukraine, sure, Russia, sure, EU ... nowp.

We now have Northern members that used to be neutral or not part of NATO, that are now part of it.

I feel like people love to misrepresent a lot of history. We have never been in a better position as a EU, vs what we used to be 40, 80, 100 years ago.

Yea, we have a lot of buildup to do again, but lets be honest, i rather see buildup now with modern kit for the modern battles, then relying on outdated 1990's doctrine and weapons. And even that is still a slow process with transitioning to the new reality of drones, drones and drones. Do not forget that 90% of the kills are now by drones.

People love to parrot those US talking point that often have no sense of history and our current EU reality in regards to security. While i admit, that we are still too reliant on US kit, even that is slowly changing. The EU moves slowly but it moves. Better then being some nations that are stuck in Imperialistic ways of thinking, like Russia.

I'd be very surprised if it wasn't more than 50% cheaper for nearly all workflows, for a very similar level of quality.

If your using pure API ... providers like neuralwatt cut that cost down even more by using energy as the actual cost. So GLM 5.2 is more expensive then GLM 5.1 on their service (those thinking tokens), compared to API costs, its dirt cheap. And way more tokens then the zai subscription delivers.

We are seeing a move towards more realistic pricing on actual consumption based usage. Be it DeepSeek, Xiaomi (MiMo), or zai's GLM via neuralwatt.

The main issue facing subscriptions a-la-carte usage, is that a lot of the heavy hitters really drain the resources. And that as a business model can not survive without ...

a) increasing the prices. b) everything goes to actual token/energy usage based billing but with more realistic pricing, and not the bloated API prices that are focused on companies.

We shall see what the future holds but things will change.

I Like Small Keyboards

I like quiet keyboards but finding non-laptop / non-membrane keyboards that do strip the leather from your wallet, is a almost impossible task. Let alone one that does not grow in noise level over time!

There is way too much focus on a entire enthusiastic click click keyboards, but everything that i keep finding, sounds like absolute horrible loud, in a quiet room, where your partner is sleeping in.

And when you find something, its often a import, not in stock, and you get presented with like a $300 bill and no guarantee about quietness after a long time using. Or worse, they changed the keys in between production runs, and its now more louder.

Why am i writing this? Don't know, maybe tired of often wasting hours seeing youtube videos and reviews, and posts about keyboards to just feeling burned out.

AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck.

If the Model makes repeated mistakes on the same subject matter, you can update your agent.md file, or you can add skills to deal with specific prompts, or you provide a better default harness.

The whole idea of coding agents is their harness makes a big difference vs a pure raw model.

However AI cannot meaningfully handle feedback and learn

How do you think models are created? They are trained on feedback and learn.

Its not cheap but you can post train models. This is how custom models are mode, that deal with specific tasks more efficiently and accurately.

Example ... Composer? Its base Kimi v2.5 model that has been post-trained 2 weeks, to create Composer 2.5, what is a much better coding model.

Its literally trained to make less mistakes by feeding it correct data. Hell, a lot of the models you are using, are often the same base model, where v2.0 was the initial released model but the model keeps training, so when they release v2.1, its still the same model, but with more training time on feedback provided to v2.0.

LLM Models are not a cake you cook one time and they are done, and you start from zero again. If you have the money, and a powerful server setup, you can take a model like GLM 5.2 and post-train it, to reduce specific errors. Sure, you need a ton of money because its a large model.

But people have been doing this with 5M, 100M, 1B, 5B models for a long time already. To the point that some of the small models can do specific tasks, almost or better then some of the huge more general trained models.

they sell you a subscription for frontier models, then over time begin to nerf them rapidly and no one talks about it.

People talk about it all the time. Just check some of the dozens of forums where its non-stop complaining about nerfs, limit nerfs, performance issues etc...

Is hard to prove that any downgrade is a effect of being deliberately served a lower class model / lower quant, or whatever. Or the "optimizations" hurting the models performance.

The TOS allows for those service "optimizations", so legally, nobody has a foot to stand upon. Like when OpenAI or was it Anthropic played with the cache, this to free up more server resources, only to later discover that its gutted the long term context behavior, and heavily degraded the models as context grew.

If you want 100% guaranteed the same performance/behavior, you need to run a model yourself (be it rented GPUs online or your own local setup). But its going to cost you a lot more ...

Care to explain what you mean by that?

why the fuck are we in the habit of giving companies the benefit of the doubt on this

Because companies pay for lobbyist, give donations, and other perks to politicians. Simple as that ... And companies a a-political, they simply give money to both sides, so who is in power has less effect.

While you the normal voter, only matter a few times, and with increased irrelevant as how powerful your vote is.

The fact that companies can now vote in local elections (in some states, you can guess the states colors), tells you how deep the corruption has gone.

If i was going to write software on this level, that will be used by governments. There is no way, its going to be a nice little program that only extract information.

Its going to have every trick in the book (and outside it), to stay hidden. And it will have payloads to alter its behavior, updates, etc...

Nobody is going to pay you big fat money envelops for software that anybody can write in a afternoon. You want it to be as capable as ever, and you do not want it found!

As far as we know that directive has remained in force with each successive administration as well.

People can state a lot, as long as your not caught.

Nothing prevent you from having the UK spy on the Germans, and feeding that intel back. Or Israel, or ... Hey, the US did not spy on a EU ally. Well, not directly and it neatly bypassed any official statements.

They might have simply gone to one of those secret court hearings and have it bypassed with a gag order in place. Officially its not done, unofficially, its been approved.

The whole "as long as you do not tell me your doing it" approach, and the politicians involve maintain deniability (even if they had the wink).

And you do not need to specific target the head off state. Plenty of side routes to still get information on meetings, that involve those heads of states. Even if your not "directly" spying on them.

So no, its a naïve way of thinking. Maybe in 20 years from now we find out, that they did spy on EU leaders. Maybe directly, maybe indirectly ... even with that directive in place. I will be amazed if they did not. Its the US we are talking about.

Fairly sure that if anybody using a advanced piece of hacking software, they are also going to delete any messages that are related to detection of such hardware.

PC viruses used to do that stuff going back so many years ago. Suppressing any notification under Windows, by disabling the AV software, its notifications, windows notifications related to it.

So it will amaze me that this is not done by any modern espionage software. Especially as the notification methods are known. Given that his device is hacked, that means a lot of avenues are under control of the espionage software. Even mails etc ... So impersonating the end user, to confirm they read a warning, is extreme easy.

I find it rather odd that people are so fixated on the idea if Kouloglou read it or not.

I think it makes houses/apartment less likely to sell. When you see the idealized version, and then the reality, the impact is much bigger then just showing reality.

Unless people prove me wrong, and they really fall for that...

Its like we used to be flooded with fisheye lens pictures of homes, that made the rooms way bigger then reality. I noticed that this trend (on the immo that i follow for years) has heavily reduced. Because nothing beats a sale, as people seeing something looking spacious on pictures and then in person seen its way more small/cramped/compact.

I love that new trend of 3D home viewing... It saves you so much time, and saves time for the immo people, filters out a lot of people with less interest.

Am i missing something? Because your making my point. Its only worth it compared to Opus 4.8, if the tasks your running requires Opus 4.8 low (or non-existing lower).

For the rest the gap in pricing vs efficiency is so small, that there is no point in using Sonnet. I am looking at their own cost comparisons vs efficiency...