Evan Armstrong | datagen/acc

Programmer, Author, ML-er. Invented Augmentoolkit which makes AI data ( I create intelligences for a living.

[Evan Armstrong | datagen/acc @e_p_armstrong]

Augmentoolkit 3.0 Released!

Augmentoolkit lets anyone make data and train AI to understand new facts or do new tasks (and more). Usage is: add files, click button. Scales well and usable for production.

- Train an LLM to understand new subjects by just adding documents.
- You can also train AI to do basically any task better just by explaining how to rate/grade attempts at that task.
- Do all this on your own hardware.
- Scales well.
- Easy to use (add files, click button).
- Running custom models works better, is cheaper, and lets you control when+how it updates.
- Contains a year and a half's worth of innovation and iteration.

https://t.co/dI5EzH9l62

[Andrew Snyder @Andrewnsnyder]

This is from C.S. Lewis's essay, "Delinquents in the Snow." I often wonder what he would have to say today.

"According to the classical political theory of this country we surrendered our right of self-protection to the State on condition that the State would protect us. Roughly, you promised not to stab your daughter's murderer on the understanding that the State would catch him and hang him. ... At present the very uncomfortable position is this: the State protects us less because it is unwilling to protect us against criminals at home and manifestly grows less and less able to protect us against foreign enemies. At the same time it demands from us more and more. We seldom had fewer rights and liberties nor more burdens: and we get less security in return. While our obligations increase their moral ground is taken away."

[Evan Armstrong | datagen/acc @e_p_armstrong]

This is quite funny. Hope the Lars is ok though. Euro govs do not sound fun right now.

[Lars Andersen @LarsAnders1620]

@Ericjduffy I published the prime minister's social security number and her phone number.

[Evan Armstrong | datagen/acc @e_p_armstrong]

I for one support @VictorTaelin's righteous crusade for midjourney against the medical establishment

[kache @yacineMTB]

LLMs be like "theoretically you could do <the best idea ever> but that wouldn't work because <idiotic slave like opinion>"

[Zy @ZyMazza]

An abliterated top-tier open weights model today can tell you how to build a nuclear weapon. The challenge to building a nuclear weapon isn't knowledge. It's not even really resources.

Iran has knowledge and resources. They can't build one.

Why? A lot of other entities with knowledge and resources have a vested interest in making sure Iran doesn't have a nuclear weapon, and they get in their way.

If you wanted to build a nuclear weapon in your basement, to vibecode one with abliterated GLM 5.2, you would find that your efforts are stymied at every turn.

Even with step by step instructions, even if it was good enough at day-trading (it's not) to make you a tidy profit to finance your operations, you would find you would quickly attract the attention of law enforcement and the intelligence community.

It turns out they really don't want you to have a nuclear weapon!

If GLM-5.2 were good enough to tell you how to build a superplague (it's not), you would quickly learn a similar lesson in biology.

It turns out the challenge to building a superplague isnt really knowledge. Just as there are thousands of nuclear physicists with the theoretical knowledge to build nuclear weapons, there are thousands of biologists who have the theoretical knowledge required to make a superplague.

But if they wanted to do it, or if you wanted to, with your abliterated GLM-5.2-Bio-Edition, you would find that you would be stymied at every turn.

First you would design your virus particle or virus-like-particle. You could do this in the computer. But now you need to make it.

Traditionally you would order it from a company like Thermo-Fischer scientific or IDT. They'll synthesize a DNA strand for you and send it back so you can inject it into a cell of your choice to get it to produce the protien you want. For a fee, they'll even get an organism to express it for you and send *that* to you.

But, it turns out they're not stupid. And if you engineer a highly virulent strand of ebola and send it to them, before they help you make that, they're gonna say "hey wait a second, this kind of looks like a highly virulent strand of ebola!"

They will get upset, law enforcement will be contacted, the cops will show up, and so on and so on.

So maybe you'll do it yourself.

You have GLM 5.2-Bio-Edition and GLM 5.2-Trading-Edition so you make a bunch of money to order the expensive, highly specialized equipment you need to do this complex synthesis at home in your basement.

So you go online with your roughly $300k and start trying to order a Mermade DNA synthesizer and assorted sundries and suddenly they're asking all these strange questions like "what's your .edu email" or "what institution are you a part of?" and you don't really have answers to those questions because you're just some guy trying to make a superplague in your basement.

And actually it turns out the Government, intelligence community, and Bio Research technologies company *also* have GLM-5.2-Supply-Risk Edition and so they notice you trying to order all the parts to make a superplague in your basement.

They get very upset and the cops get called and so on and so on.

And *actually* it turned out *everyone* had access to GLM-5.2 Trading edition, so it wasn't so easy to make a killing on the market to finance your operation in the first place because your agent wasn't placing trades against human rubes, but against a whole hoard of *other* agents every bit as intelligent as itself, so the net result is that the market became slightly more efficient, but no one was really able to lock in asymmetrical gains.

Unless, however, *one* person had access to the GLM-5.2-ASI-Edition. Because if that person had access, they would be trading against rubes-by-comparison, and they would make a killing in the market to finance whatever operations they please.

And their ASI would be smarter than anything anyone else had, so it could engineer a VLP that thermo-fischer wouldn't be able to detect and reject on site. Or it would be able to deftly bypass supply chain safeguards, hacking into MIT or something (who has no ASI defender) to get you a .edu email and the appearance of institutional backing.

See, in a world where everyone has ASI, it's as though no one does. It's just normal society, where everyone's wishes and capabilities are mediated by group dynamics, everyone kind of has to stay within the overton window, and deviations are policed largely by the community, and then by the state when need be.

But in a world where only anthropic has ASI, or only openAI, or only the US government, then the *first* bad idea they have is immediately implemented with no resistance, with no hope of stopping it.

And on a long enough timeline, its only a matter of when, not if, they will have a bad idea.

Pluralism has served America well for a long time, and when it comes to ASI, I'm a pluralist too.

[Evan Armstrong | datagen/acc @e_p_armstrong]

@Italofiend Please don't do this.

There are better things ahead and,
"For ye are bought with a price: therefore glorify God in your body, and in your spirit, which are God's."
God wants you here and so do we.

[Will Solfiac @willsolfiac]

Man fraudulently applies for PIP using TikTok guides to prove it won't work - gets awarded PIP.

This subreddit is on a roll recently @LegalYookay

[Taelin @VictorTaelin]

In decision theory, the value of information is always non-negative for a rational agent. Extra information only hurts when a process uses it sub-optimally (overfitting, being misled by noise).

So these mastermind doctors see a result proving their decision making is utterly broken, and interpret it as more information being bad. Incredible. Absolute genius.

We make fun of our antecedents for using uranium watches or leaded gasoline, only to do shit like this. We will be laughed at so hard it makes me intellectually embarrassed to live in this period of time.

[Roman Helmet Guy @romanhelmetguy]

The ancient Britons burned down London after the Romans raped two women. Today's Brits will do nothing. Not because they are more enlightened, not because they are more moral. But because they are less.

[Aditya Mishra @adityamtweets]

@Fortune21Good He can choose to give a bonus out of good faith. But no one should feel entitled to it.

[Evan Armstrong | datagen/acc @e_p_armstrong]

@Ian_Fisch @Mystikart_ The advice is uncomfortable but you're onto something. I actually think it looks very nice, but the gameplay is just... I guess it's too artsy for me? idk

[Lefrizee 🇺🇸 @lefrizee]

URGENT: my dad is still in the ICU and his condition is worsening. It’s not looking very good and I’m not sure if he’s going to make it. PLEASE PRAY

[Taelin @VictorTaelin]

this is a weird long post without much substance

I strongly recommend against reading it

...

so, do you feel like whatever you're working on right now is pointless, or will have zero value soon, due to the crazy times we're living? then, perhaps you should stop, and start working on the only unsolved problem that actually matters TODAY:

✨ replicating GPT-3 in a laptop ✨

"why is that so important?"

because it would make AI incredibly cheap, which would mean everyone would have Fable-class models in their laptops, without depending on Anthropic, OpenAI, or any other hyper-scaler giant. and that's amazing, don't you think?

"isn't that literally impossible?"

that's the cool part: as far as computer science is concerned, no. not really. not at all. is entirely plausible and, as far as we know, most likely not even hard.

it takes one good idea. one breakthrough. one great "aha moment", to go from zero to "hey, this software I wrote is producing credible English sentences"

and whenever that happens:
- the entire AI industry collapses
- clusters are liquidated
- we all get Fable at home
- you become famous and rich, if that's your thing

sounds fun, doesn't it?

"wtf you talking, OF COURSE that is hard"

so prove it.

show me a paper, a lean file, anything that proves that training a Fable-class model fundamentally requires billions of dollars. you can't, because, guess what - it is not true! the only "evidence" we have is purely psychological. "many attempted over decades, and the best thing we have is GPTs, so, it is a hard problem" - but that's not a scientific argument. that's a human, psychological, sociological argument. and if that's it, consider the following counter-argument:

✨ humans are stupid as hell ✨

I mean, 10 years ago we didn't have transformers, so, that very argument could be used against GPTs existing. yet, they exist. we have them now, because someone found it. and, guess what, it isn't even complex. I mean, karpathy implemented the whole thing in a napkin. and it probably compiles.

we were just too dumb to figure GPTs out... for decades.

just like GPTs, there ARE other approaches, other algorithms, other architectures, equally simpler or even simpler, that do work. this is a mathematical certainty. and one of them might be astronomically faster than what we're doing right now.

and you might be the one to find it!

"me? why me???"

because you're intelligent, creative and handsome.
I see a lot of potential in you.
in fact, I always believed in you.

and I think you're wasting your time, doing that silly agent orchestrator. nobody wants that. quit it. take your most interesting ideas, intuition, creativity, and work in a problem that matters. do your best shot at reproducing GPT-3 in your own laptop.

do NOT fork llama.cpp.
do NOT train another LLM.

do something... ✨different✨

it must be unique, novel, full of YOUR soul. something nobody thought of, or bothered doing.

go ahead and implement that thing in C/CUDA (or Bend!).

no Python!

zero excuses for Python.

any model is fluent in GPGPU now. build a real kernel.

and then, train your thing. download wikipedia, give it time and compute to absorb the patterns of English speech. you can rent GPUs anywhere nowadays. let it train. then, ask it some questions. chances are it will just respond back. just like GPT-2 answered OpenAI. computers are incredible. don't underestimate them!

"many tried. nobody succeeded. why would I?*

see - that's your mistake again. turns out not many actually tried, at all. I promise you. who do you think is seriously working on that?

people on Mozilla?

they're busy building a browser

Linus Torvalds?

he is busy building an OS

employees at OpenAI, Anthropic, xAI?

they're paid to work on what is proven to work: GPTs.

what about all the AI enthusiasts all around the world?

yeah, you know they're mostly fine tuning Qwen

and how about your friends?

if only they weren't busy building a SaaS in the eve of AGI...

how about people from the past?

bro - people from the past seriously expected Lisp would be AGI. just dismiss them. they didn't have the compute, the resources, the knowledge, the MODELS that we have today. that YOU have access to.

so, what's left? not much.
the world looks big. it is not.
truth is: ✨almost nobody is working on this ✨

"I still think it is impossible. I don't trust you"

well, take my word no more.

Ilya himself, in his 2019 talk on GPT-2, said:

> "the story of deep learning is this: empirically old simple methods which were usually invented in the 80s and the 90s when scaled up on very large clusters work really well."

and then:

> "(we took) normal simple reinforcement learning method, scaled it up, and discovered that it suddenly becomes very capable of solving extremely hard problems."

and again:

> "you take a simple tool which is unimposing and barely works, and then you run it on a big cluster and suddenly it works, it becomes a capable tool for solving problems"

do you see the point here?

Ilya isn't arguing that transformers are magic.
Ilya is arguing that SCALING is magic

step #1: take a simple, elegant algorithm.
step #2: shove compute at its face.
step #3: ...?
step #4: your computer is talking to you

THAT is the key insight that led to GPT-3

THAT is what Ilya saw

THAT is what caused the OpenAI x Anthropic war

THAT is the founding principle of the ongoing era

not "scaling transformers work"

but "scaling beautiful algorithms works"

that's the incredible lesson.

yet, we all took it and... threw it way.

- zurk bought 100k GPUs. to train GPTs

- musk bought 100k GPUs. to train GPTs

- bezos bought 100k GPUs. to train GPTs

...

that's what everyone is doing.

so, no. not many are trying to replicate GPT-3 through other means.

we're just ants, after all...

whenever we find a pile of sugar, we leave a track of pheromones, which guide the rest of the colony towards the new food source. the colony then swarms around the pile, extract all of it, until no grain is left.

but piles of sugar aren't spontaneously generated in the middle of nowhere. they imply something more profound: "humans are around". and, if humans are in sight, even better things must be. like a big sweet cake.

a colony that only follows the pheromone trail would miss the cake for the grains. that's why every ant species has scouts and exploratory foragers. and, just like a pile of sugar implies something more profound, LLMs also imply something quite profound:

*computers are capable of thinking*

a pile of sugar is never alone.

GPTs are most likely not the only system capable of thinking.

so, if you find yourself a bit lost, without purpose, like your work is pointless and Fable 3 will soon one shot it anyway... consider becoming a scout. find a new approach to AI. bring something new to humanity. breaking out of the massive cost associated with training GPTs is the next big step in AI, and it will only happen if people like you work to make it happen.

[woods @wuooods]

anyone who wants to be the guy on the right needs to understand the guy on the left is keeping your toys from becoming cheap garbage that suck to play with

[IMPERATOR @IMPERATORAUS]

To act bravely is the part of a Roman, to suffer bravely is the part of a Christian.

– St Ambrose (Commentary on the Beatitudes)

[Trad West @trad_west_]

>Plot centers entirely around fighting extreme government taxation and overreach
>The Sheriff of Nottingham literally collects taxes from the church poor box
>Friar Tuck gets so fed up with the state disrespecting the Church that he physically throws hands with the Sheriff
>Casts the Crusades in a positive light
>Male protagonist who risks his life for his people
>Unapologetically traditional romance with Maid Marian without any modern subversion
>Climax is literally a raid to break political prisoners out of a corrupt jail
>Story resolves when the rightful, divinely-appointed monarch returns from the Holy Land to crush the corrupt politicians
>Ends with a beautiful church wedding and a happily ever after

We need to make Kid's stories based again

[furnaces @furnaces]

@ComedicBizman locking yourself in a dark room for 30 minutes will do better than any self help video.

[Tej Dosa @ComedicBizman]

The fastest way to turn into a NPC is to fill every moment of stillness with audio books, podcasts, CEO interviews, tweets, threads, and YouTube videos.

The fastest way to turn into the Main Character is to spend more time in stillness and give yourself 4 hours to create.

[Six: Sotto-Capo of the Thumb 🇮🇩🇺🇸☭⃠ @Six_senseNeo]

@Jek_Korpen One thing I've accepted is that

"Players will optimize the fun outta games."