@zeddotdev Cool. I hope there will be improvements to how Mermaid renders in Markdown files down the road, too.
English
Arthur Mason
349 posts
Arthur Mason
@ArthurLeeMason
AI Realist | Writing The Will to Game | Gamedev @CoolestCatsLab with my beautiful wife @AnyaLeeMason
Katılım Mart 2022
99 Takip Edilen23 Takipçiler
Yes, leadership is an art and a vital one in more complex undertakings. A ship is best with a competent captain, an orchestra with a competent conductor, an army with a competent general. Strong leadership is less about judging people as either "good" or "bad" and more about judging people as having distinct aptitudes and being able to marshal those differences toward a single goal.
内藤 時浩@NAITOTokihiro
仕事がデキるってのは、個人の能力と思われがちだが、違う能力もあります。それは、仕事をデキる人を集める能力です。仮に本人は何も出来ないとしても、デキる人が集えば何だってデキる。そして、今の時代、案外そういう人が望まれている風潮すら感じる。 学校の卒業制作とかで、なんであいつの周りにあんな凄いやつばかり集まっとんねんっていう疑問、感じたことはないですか？
English
I've been wrestling with the same problem. The BYOM (Bring Your Own Model) approach offloads the cost onto the players, but few players will tolerate this. You can run small local models on-device to avoid fees entirely, but this means players need better hardware, and weaker models constrain what you can design.
The approach I keep returning to is also the most difficult, but I think it'd be worth it if you pull it off: design as much of your system around deterministic pipelines as you can while reserving heavily quantized, fine-tuned local models for only the outputs that genuinely require non-determinism. For this approach, I've been exploring graph topologies, since they can give the deterministic layer more depth without significantly increasing hardware cost relative to inference.
English
@fchollet Regulations are only counter-productive for industries in the short-term. This is because regulation is about societies, not industries. Society is the substrate upon which industries are built, however.
English
Even if you are in favor of AI regulation, you should recognize that opaque and arbitrary regulatory strikes are counter-productive for the whole industry.
English
@leon_devops This is partly because AI has brought in a wave of non-engineers into the conversation, driving some engineers into isolation. But it doesn't need to be this way. New communities are also forming around AI, and some of them are just as passionate about technology.
English
Any learned competency will degrade over time if it isn't occasionally reinforced through exercise, including our capacity to reason through problems. However, the competencies we learned very early in life or inherited from our parents aren't at risk, because they don't unravel so easily.
English
AIに頼ることで、人間がバカになっているのではなく、AIの圧倒的な性能によって、相対的に人間の無能さが露呈しているだけ説。
数十年、自分の頭を使って生きてきた人が、たかだか1年、AIを使ったからって、急激にバカになるはずがない。
日本語
@igz0 I don't think engineers have changed much. I think coding models have simply brought in more non-engineers into the conversation. If you explore a platform like Leanpub, you'll still find engineers experimenting with custom languages, databases, and the like.
English
ITエンジニア、昔は「俺の考えた最強の自作言語」みたいなの作っている変態が多くて面白い人種だったのに
今は一日中「Claude Codeの開発者による生産性向上のベスト・プラクティス」みたいのを拡散するクソおもんな人種に成り下がって悲しい。
日本語
There are two camps among AI enthusiasts.
The larger camp seeks a freedom-from. They yearn for redemption from the "permanent underclass" through AI. They are Nietzsche's slave moralists in the clothing of accelerationism. Unsurprisingly, this is also the camp concerned with "AI consciousness" and "AGI" — because the slave always needs a master to kneel before.
The smaller camp, in contrast, is driven by a freedom-to. Not salvation, but enablement, is what concerns them. To them, the machine wills nothing; it is merely an instrument for carrying out their goals and creating what is already within them. The will of the ancient masters flows in them.
#GenAI #philosophy #Nietzsche
English
Yes, and this is related to why so many who are against AI ask, "What is there to even do with it?"
Those who ask that question had no backlog of creative projects when neural networks became viable as tools. There was nothing waiting inside them, so when the technology arrived, they either felt confused, or they sensed the future that was coming — the future that they clearly won't be part of — and felt threatened.
AI is for the technologically modern generation of creatives, those deep in the new global culture of digital technology and the internet, who have spent decades growing their ideas and refusing to compromise on them, because the means to authentically realize them didn't yet exist.
To hearken back to Nietzsche's famous passage: new technologies, regardless of the era they emerge in, always reveal who still has the chaos of creativity in them, and who has none.
#GenAI #creativity #philosophy
とむい@tomuisan
生涯で蓄積されてきた、 「やってみたかったけど技術的、時間的、コスト的に難しくてできなかったこと」 がAIによっていきなり可能になっちゃったから今忙しいんだなって思った。
English
I understand how neural networks are built and their limitations. And yet, despite knowing that what they promise won't be fully delivered, I still cherish seeing advertisements for new models coming out, for they reflect the deepest and most innocent part of our psyche.
All children dream of being able to create whatever is in their imagination. Desires for status and safety are alien to them; these are things for those who have abandoned the child within them.
Heraclitus: "Time is a child playing dice; the kingly power is a child's."
Generative AI may bring with it enormous economic and social disruption, but the child in humanity desires it. It wants a tool that can create anything it can imagine, so it can play with the imagination unrestrained.
#GenAI #philosophy #Heraclitus
English
If I recall correctly, there was no mention of any HIPAA compliance anywhere on their website a year ago. It's there now, but it wasn't for any of the key providers. The reason for that is because they've all been using user prompts for QA and training. Remember that these are corporate products, not community services.
English
My team loves Claude from @AnthropicAI . But this new policy of retaining prompts and usage is a red line...we simply can't give over our usage. Prompts contain our IP; literally all our design files and docs. Why would this ever have been ok?
It's sad because everyone was looking forward to using the new model. Sigh.
English
I don't see why this would be a problem, should things move in this direction. Model specialization is beneficial in numerous ways.
First of all, most people aren't using AI beyond their own specialization, so they don't need one massive generalist model that can do it all. They just need a specialist model that can do whatever they care about very well and very fast.
Secondly, compartmentalizing specialties into separate models makes the product management and QA pipelines more efficient, because now each model has a specific target audience to gather feedback from and optimize for.
Thirdly, this compartmentalization allows markets to self-regulate more efficiently. If the legal AI is getting 10x more attention than the biology AI, then supply will adjust to meet that demand hierarchy accordingly.
All of these contribute to shipping better products. This is why we even *have* specialization as a concept, because it supports the development of better products and a higher quality of life overall.
Your complaint boils down to not being able to receive all of these services for one low cost anymore, despite that pricing structure being unsustainable and unsupportive towards the development of better products.
Kirk Patrick Miller@Chaos2Cured
You all want to know the truth? Dario and Anthropic are planning to release individual models for different things. Not because AI can’t do biology (lie), but so they can sell you - a law AI - a medical Ai -a chemical AI -a biology AI… And they will only allow certain people with licenses to use it. None of this is hidden. Look at Microsoft. They announced individualized Ai. If they do this, the universities backing them get to determine who gets to ask what. Think about it. Last year I was able to design new molecules. Now, it is blocked. Last year the AI could simulate biological systems to design treatments and cures. Now it can’t. If the AI are improving, why are their abilities vanishing? MONEY AND POWER… Anthropic just has better marketing than OpenAI. They are doing the same. And this is the real reason I support xAI as they too do things I dislike. I am begging anyone who sees this to think about what I am stating. The top 1% are planning to take it all away and force you to play their game. Fight for truth, transparency, and equal access. With everything you are. This is the end game. •
English
Recursive self-improvement, considered the embryo of artificial general intelligence, is meaningful insofar as "improvement" is meaningful, which is contingent on the body. What is "improvement" to a data center? It isn't "to better understand" or "to feel more," but something colder, something closer to the electrical charges beneath the 1s and 0s.
#AGI #RSI #philosophy
English
The people use words like "thinking," "reasoning," "learning," and "agency" to describe processes related to neural networks — because they lack the eye of a psychologist.
#neuralnetworks #psychology
English
The next great event in art will emerge from a people and region that has previously never enjoyed the limelight. AI is the most powerful creative tool ever built, but only the right culture will make proper use of it — a culture that's only just beginning to form.
The loudest voices against it aren't defending art. They're defending their rank in modern society. Their pessimism is just masked fear. They can't see the future because the past already has them in its clutches.
And it was always like this. Every birth of a culture provokes the panic of those it renders obsolete.
#GenAI #philosophy #Nietzsche
English
Arthur Mason retweetledi
「AI には意識があるのか？」という議論は「人間の脳には意識がある」が前提だけど、僕はこの前提が間違いであり、「地球は宇宙の中心である」「人類は進化の頂点である」系の錯覚だと思ってる。我々は特別だという願望から空想を生み出し、いずれ科学の進歩によって否定される
日本語
ASI is often framed as an ultimate controller, something that will command and rule over the world in the future. This belief rests on a profound misconception.
A thermostat does not "control" temperature any more than gravity "controls" the orbit of a planet. In both cases, we are projecting intention onto a process. Control, in the authentic sense (etymologically, the word stems from the action of rolling a wheel), is an action belonging only to something that can experience orientation. Orientation does not originate in computers; every "sense of direction" a computer possesses was given to it. So it is a misstep to attribute to them any capacity for control.
#philosophy #AI #ASI
English
Creativity is at an all time low in the world. At the same time, the most powerful creative tools ever invented are being engineered. Is this paradoxical?
No. It always works precisely like that, in fact. This is because it takes a long time to create something truly new, and every creation involves exhausting the resources of another. Every child begins as a wholly dependent life within the mother, drawing directly from her vital reserves and strength to sustain its own growth until birth. Ideas exert this same internal demand, consuming one's focus and energy until they are finally given form.
Therefore, it is perfectly normal for society to both evolve and degenerate at the same time. Tomorrow will be something new. Right now, the situation may look bleak, but only because we are not the inheritors of this new tomorrow.
#creativity #philosophy
English
Machines don't judge; they average.
Judgment requires exclusion, negation, rank, and distinction. It isn't merely a computational operation. It's a libidinal pull towards a good and a bad, a yes and a no. We call the refinement of that pull "taste," and the source of all taste is the will to power. Neural networks have no libido, no will to power, and so nothing to judge with. Left to themselves, they drift toward the mean rather than meaning.
Context engineering is therefore the imposition of taste upon a machine that has none. But no matter how elaborate our prompts, skills files, and examples become, the average creeps back in wherever our blind spots allow it.
#Nietzsche #contextengineering
English