Import AI 474：柏拉图式心智空间；太空 TPU；智谱开启外层 RSI 循环

欢迎来到 Import AI，这是一份关注 AI 研究的通讯。Import AI 依靠 arXiv、卡布奇诺，以及读者反馈运行。如果你愿意支持这份工作，请订阅。

心智是否是来自柏拉图空间的模式，而身体与机器只是它们的接口？Michael Levin 如此发问：
……一篇令人脑洞大开的论文，要求我们重新审视哲学与生物学的基本假设……
我们周围世界中的复杂性，是否可以更好地解释为：存在一个具有任意复杂度模式的柏拉图区域，而这些模式试图将自身嵌入世界？这正是科学家 Michael Levin 一篇离经叛道论文的核心大问题。

论点：“我认为，新兴的合成形态学和多样智能科学表明，心智的非物理主义模型是可能的，并展示了如何对其进行实证研究……我提出，心智与大脑之间的关系，与数学模式及其所引导的形态发生结果之间的关系相同（更广泛地说，心智:身体 正如 数学:物理）。”他写道。“身体（无论是活体、工程化实体还是混合体）都是一个庞大、多尺度模式层级进入物理世界的接口……由生物物理、生物力学或化学信息场表征的特定模式，可作为稳态或异稳态过程的编码设定点——换言之，即系统试图在各种问题空间中导航至的目标状态。”

走向由任意复杂模式构成的柏拉图领域：“我们已经知道，非物理模式会进入非生命与生命世界，并在功能上产生重要影响；而且我们能够（并且确实）卓有成效地研究它们……柏拉图形式会向物理事件注入信息并施加影响，例如影响生物体的生长与形态。”他写道。“这一潜在空间[柏拉图形式的空间]不仅包含整数事实、几何形状等低能动性形式，也包含范围广泛、能动性不断增强的模式，其中一些我们称为‘不同种类的心智’。因此，我提出，心智作为使躯体具备灵魂的模式，与那些栖居于并引导简单物理结构行为的模式，在非物理本性上属于完全相同的类别（但程度不同）。”

**一些奇异的实验：**目前并没有能够一锤定音地支持这一观点的实验；相反，世界中存在一些怪异现象，暗示即使看似简单的系统，也能够探索并扩展到丰富的模式之中。这些模式在假定的复杂性柏拉图空间内似乎与它们相邻，却并非由系统自身所蕴含。

这里有两个不错的例子：异种机器人（xenobots）由青蛙细胞制成，是一种生物机器人，它们“让我们了解与青蛙胚胎模式相邻的模式”；而人类机器人（anthrobots）由人类气管细胞制成，在离开人体后会呈现怪异的新形态和有趣行为，例如能够自主修复神经元损伤，并且“让我们了解与成年人体组织相邻的模式”。

文中还精彩讨论了一项实验：研究者采用标准排序算法，并以几种方式对其进行扰动，包括将某些单元格“锁定”在原位，随后发现算法会绕开这些单元格。另一项更复杂的实验中，他们让每个数字运行不同类型的排序算法，然后研究不同排序算法家族如何在排序过程中出现，并在数字空间中聚集。仅从抽象角度审视这些算法，无法解释或预见这类复杂性；但当它们嵌入复杂系统时，便突然呈现出新的属性。

“机器（无论是肉体机器还是硅基机器）也会做一些并不在算法中的事情，我们也是如此；而这些事情不只是不可预测的复杂性，它们是智能以及心智的其他组成部分。正是这些行为——算法所允许、却并未直接规定的行为——对应着我们在试图理解自由心智如何在生命体中涌现于化学决定的基质之上时，所寻求的自由（物理上非决定论的）或秘密配方。”他写道。“在这一视角下，算法机器和生化生命完全处于同一光谱上：它们都具备超越物理或算法实现事实的能力，因为二者都只是通向某些模式的指针或接口；这些模式以一种让我们获得多于投入的方式进入世界。”

**这为何重要——AI 与人类的“心智”或许在柏拉图式心智空间中彼此接近，但并不完全相同：**Levin 提出的研究议程令人沉醉、引人入胜；在我看来，它似乎也为我们如何处理 AI 系统引发的深层哲学与对齐问题提供了一些线索。也许，人类心智和不同类型的 AI 心智是这一空间中相邻的形式，而大脑和数据中心则是它们的投影所落在物理世界中的不同生物系统或锚定系统？

我们会不会只是被来自别处的模式殖民的容器？是否“可以将进化论重新表述为一个由具备能动性的模式寻求实体化身的过程”？

“除了物理对象的‘只要建造，它们就会到来’模型之外，是否存在某种‘力量’，从这一空间中拉取模式？又或者，柏拉图空间的内容是否处于某种‘正压’之下，以某种方式鼓励它们作为侵入性想法、原型、艺术作品而在世界中出现？……目前唯一能够明确说的是：我们对于物质能力的无知，以及对于进入特定架构的模式的无知，仍然极其巨大。”

给自己泡一大桶咖啡，花几个小时读读这篇论文——很值得。

延伸阅读：《进入心智：因果性、非物理模式为自然、合成与混合实体赋形》（MDPI）。

---

机器人技术正在为 LLM 时刻做准备，但它缺少一种通用算法：
……要让机器人走向成熟，后训练必须变得更简单……
斯坦福研究员 Perry Dong 与斯坦福教授、机器人公司 Physical Intelligence 联合创始人 Chelsea Finn 撰写了一篇很好的文章，讨论了制约机器人发展的因素。

**LLM 与机器人：**对于 LLM，事实证明，重要的是“该领域逐渐形成了一套共享的语言模型后训练配方”。他们写道，这一配方基本包含四个步骤：1）拥有强大的预训练模型；2）定义环境与奖励机制（例如偏好模型）；3）针对参考模型运行 RL 优化；4）监测并解决奖励黑客等病理问题。

“机器人技术所处的位置，几乎正是语言建模当年所处的位置：预训练已经实现了出色的规模化。”他们写道。“缺失的是另一半：模型从自身经验中学习。为此，我们需要一套后训练配方。而对机器人而言，它需要比语言模型更可靠。”

**机器人需要什么？**他们写道，需要“一种专门为微调前沿机器人模型构建的算法，它在应用于数十亿参数模型时保持稳定，并且能从足够少的经验中学习，从而在真实硬件上具备实用性”。此外，行业还需要就“围绕该算法的一套标准实践”达成共识：一套定义何为成功的默认方法；一套在每次尝试之间重置场景、让机器人能够再次尝试的默认方法；以及一套让人类提供反馈、并将反馈转化为学习的默认方法。

在文章中，研究者讨论了一种他们正在为机器人开发的特定算法 EXPO(-FT)，他们认为它具备其中一些特质，尽管它仍处于早期开发阶段，尚未被广泛使用。“EXPO(-FT) 的工作方式是：利用轻量级策略的小幅修改，通过强化学习反复改进前沿模型的动作，然后将这些改进吸收到前沿模型自身之中。”

**这为何重要——标准配方是大规模扩展的前提条件：**专有的大规模 LLM 已经足够智能，可以向机器人下达指令，并帮助其在现实世界中构建和执行复杂计划；但机器人的实际移动和视觉能力仍相当初级。解决这一问题将需要围绕机器人运动与视觉定制的模型，而这又需要上述标准化工作。

“语言模型后训练之所以能够规模化，是因为该领域确立了足够具体、可遵循且预期有效的默认方法。机器人技术正在来到同一时刻。”他们写道。“就一套行业默认标准、一种通用后训练配方达成共识，是让我们走到那一步最重要的部分。”

延伸阅读：迈向机器人的通用后训练（Perry Dong 博客）。

---

太空计算机！我再说一遍：太空计算机！
……Google 准备将 TPU 送入太空……
Google 更新了 Project Suncatcher 的进展。这项计划于去年公布，旨在将计算机送入太空，并最终在那里训练 AI 系统（Import AI #434）。Google 正与合作伙伴 Planet 一起，准备通过 SpaceX 的“Transporter-18 拼单发射任务”，将部分芯片送往太空。

**为太空做准备：**Google 已对其 TPU 进行了压力测试，以检验它们能否承受进入太空时巨大的 g 力。Google 还测试了它们对辐射的响应，发现其 Trillium TPU“表现出极强的耐受性，能够承受超过五年太空任务期间所接收总电离辐射剂量的辐射”。目前，Google 正研究太空散热问题，这可能是一项挑战：计算机芯片会产生大量热量，而在真空中通过辐射散去这些热量极其困难。

**这为何重要——AI 训练和 AI 推理的未来将离开地球：**尽管 Suncatcher 听起来像科幻项目，但只要考虑到 AI 所用算力及其能源需求持续增长，它实际上很合理。什么地方既拥有巨大空间，又有优质太阳能？太空。我认为，人类很可能会迅速将大量计算迁移到轨道中，尤其是在奇点驱动的自动化覆盖 AI 供应链之后。

延伸阅读：Project Suncatcher 背后：我们将 AI 送入太空的登月计划（Google 博客）。

---

智谱利用 GLM-5.3 自动化部分自身基础设施工作：
……中国开发者开始进入外层 RSI 循环……
中国公司智谱 AI 是 GLM-5.3 的开发方，GLM-5.3 是全球最强的开源权重 LLM 之一。该公司发表了一篇文章，介绍它如何使用自家模型帮助建设自身基础设施，让我们得以窥见 AI 实验室如何试图利用自己开发的技术来提高研发速度。

具体而言，这篇文章详细说明了他们如何利用该技术帮助发布 GLM-5.3 Flash——这是其最强模型的快速、低成本版本。

他们做了什么：“GLM-5.3-Flash 的发布建立了一个由工程师、Infra Agent 和实验环境组成的优化循环。工程师定义目标和系统边界；Agent 负责分析、提出假设和修改代码；实验环境则提供分层、及时且可验证的反馈。”他们写道。“大量工作由 GLM-5.3 驱动的 Infra Agent 完成……随着 Infra Agent 的反馈循环贯穿整个优化过程，GLM-5.3-Flash 从初始模型适配到具备生产就绪能力用时不到两周，最终端到端吞吐量相较初始基线提升三倍。”

**如何自动化你自己：**该公司分享了让软件借助 AI 实现更高程度自动化的建议。

• 反馈必须足够局部化：“它应与特定的引擎启动参数、代码改动、内核、输入条件、线程、执行区间或代码路径相关联，从而帮助 Agent 缩小问题范围。”

• 获取反馈的成本必须低且足够及时：“更短的验证周期可帮助 Agent 及时纠偏，并减少在无效假设上投入的精力。”

• 反馈必须支持客观验证：“某项修改是否正确、性能是否有所提升，应通过参考实现、测试结果和可比较的实验指标来判断。”

**这为何重要——实验室提速的迹象无处不在，而人们对此感到矛盾：**这里，我们看到一家 AI 实验室使用自己最强大的模型来加速基础设施及其他模型的开发。这类故事过去非常罕见，如今却越来越常见，并且顺应了 AI 系统能力持续增强这一更广泛趋势：“在 GLM-4.7 之前，我们内部使用 GLM 编程时带有一定程度的义务感。毕竟，它是我们自己的作品。”他们写道。“如今，GLM-5.3 已成为团队中每个人不可或缺的日常编程伙伴，并且正稳步走向取代我们。”

文章后半部分，他们以更为伤感的语调谈及 AI 开发正在发生的更广泛变化，以及人类目前的比较优势在于做出更高层级的设计决策，尽管这种优势也可能很快消退。“我们认为，人类应当在未来很长一段时间继续守住这条界线。”他们写道。“这一边界上的进展不会仅仅因为我们希望它放缓而放缓。”

延伸阅读：[迈向递归

Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If you’d like to support this, please subscribe.

**Are minds patterns from a Platonic space, with bodies and machines as their interfaces, Michael Levin asks:
**_…A mind-bending paper asking us to reconsider basic assumptions about philosophy and biology…
_ Could the complexity in the world around us be better explained by there being a platonic zone of patterns of arbitrary complexity which seek to embed themselves in the world? That’s the big question at the heart of an iconoclastic paper from scientist Michael Levin.

The argument: “I argue that the emerging sciences of synthetic morphology and diverse intelligence suggest non-physicalist models of mind and show how they can be empirically investigated… I propose that the relationship between mind and brain is the same as the relationship between mathematical patterns and the morphogenetic outcomes they guide (more broadly, mind:body is as math:physics)”, he writes. “Bodies (whether living, engineered, or hybrid) are interfaces for a massive, multi-scale hierarchy of patterns to ingress into the physical world… specific patterns represented by biophysical, biomechanical, or chemical information fields serve as encoded setpoints for homeostatic or allostatic processes—in other words, goal states toward which systems try to navigate in various problem spaces”.

Towards a platonic realm of arbitrarily sophisticated patterns: “We already know that non-physical patterns ingress into, and functionally matter, in the non-living and living world and that we can (and do) study them to great effect… platonic forms inject information and influence into physical events, such as the growth and form of biological bodies,” he writes. “This latent space [of platonic forms] contains not only low-agency forms such as facts about integers and geometric shapes, but also a wide range of increasingly high-agency patterns, some of which we call ‘kinds of minds’. Thus, I propose that minds, as patterns that ensoul somatic embodiments, are of exactly the kind (but not in degree) of non-physical nature as the patterns that inhabit and guide the behavior of simple physical structures.”

Some bizarre experiments: There aren’t any home-run experiments that back this up, but rather there are weird things in the world that suggests that even seemingly simple systems are able to explore and expand into rich patterns that seem to be adjacent to them in this hypothesized platonic space of complexity, yet not implied by the systems themselves.
A couple of nice examples here are xenobots, which are biorobots made of frog cells and “teach us about patterns adjacent to those of frog embryos”, and anthrobots which are made from human tracheal cells and when taken out of the body take on bizarre new forms and display interesting behaviors like being able to autonomously heal damage to neurons and “which teach us about patterns adjacent to adult human tissues”.
There’s also an amazing discussion of an experiment where they take a standard sorting algorithm and perturb it in a few ways, including ‘locking’ certain cells in place and noting that it then routes around those cells, and a more complex experiment where they let each number run a different type of sorting algorithm and then study how different families of sorting algorithm appear and then cluster together in numberspace during the sort - this kind of complexity is unexplained and not anticipated by staring at these algorithms in the abstract, but by embedding them in a complex system they suddenly take on new properties.
“Machines (whether meaty or silicon-based) also do other things that are not in the algorithm, as do we, and these things are not just unpredictable complexity, it is intelligence and other components of minds. It is those behaviors—allowed by the algorithm but not directly prescribed by it—that correspond to the freedom (physically non-determined) or secret sauce that we seek when trying to understand how free minds can supervene on chemically determined substrates in the case of living beings,” he writes. “On this view, algorithmic machines and biochemical life are on exactly the same spectrum, having in common the ability to go beyond the facts of physical or algorithmic implementation because both are just pointers/interfaces to patterns that ingress in a way that results in getting more out than we put in.”

Why this matters - perhaps the ‘minds’ of AI and of humans are close but not exactly the same in platonic mindspace : The research agenda Levin lays out is intoxicating and fascinating and also seems to me to hold some clues as to how we might grapple with the deep philosophical and alignment questions inspired by AI systems. Perhaps both human minds and different types of AI minds are neighboring forms in this space, and brains and datacenters are different bio- or anchor-systems in the physical world that their shadows fall on?
Could we be but vessels colonized by patterns from another place? Could it be “possible to re-cast the theory of evolution as a process in which agential patterns seek embodiments”?
“Is there a “force”, beyond the “if you build it, they will come” model of physical objects, pulling patterns from the space? Or are the contents of the Platonic space under “positive pressure”, somehow encouraging their appearance in the world as intrusive thoughts, archetypes, works of art?... The only thing that can be said strongly at this point is that our ignorance about the capabilities of matter, together with the patterns that ingress into specific architectures, is vast”.
Make yourself a bucket of coffee and spend a few hours with this paper - it’s worth it.
Read more: Ingressing Minds: Causal, Non-Physical Patterns In-Form Natural, Synthetic, and Hybrid Embodiments (MDPI).

---

**Robotics is preparing for a LLM-moment, but it’s missing a universal algorithm:
**_…For robots to mature, post-training needs to simplify…
_ Perry Dong, a Stanford researcher, and Chelsea Finn, a Stanford professor and co-founder of robot company Physical Intelligence, have written a nice piece about what is holding back robotics.

LLMs vs Robots: For LLMs, what turned out to be important was that “the field converged on a shared recipe for post-training language models”, which basically had four steps: 1) have a strong pretrained model, 2) define the environments and reward (e.g., preference models), 3) run RL optimization against the reference model, and 4) watch for and address pathologies like reward hacking, they write.
“Robotics is sitting almost exactly where language modeling was: The pretraining has scaled beautifully,” they write. “What is missing is the other half: the model learning from its own experience, and for that, we need a recipe for post-training. And for robotics, it needs to be even more reliable than language models.”

What does robotics need? “An algorithm built specifically for fine-tuning frontier robotics models, one that stays stable when applied to models with billions of parameters, and that learns from a small enough amount of experience to be practical on real hardware”, they write. Along with this, they need to converge on “a set of standard practices surrounding the algorithm. A default way to define what counts as success. A default way to reset the scene between attempts, so the robot can try again. A default way for a person to give feedback, and to turn that feedback into learning”.
In the post, the researchers talk about a specific algorithm, EXPO(-FT), that they’ve been developing for robots which they think embodies some of these qualities, though it’s early in its development and not widely used. “EXPO(-FT) works by learning to repeatedly improve actions from the frontier model using reinforcement learning with small edits from a lightweight policy, and then absorbing that into the frontier model itself.”

Why this matters - standard recipes are a prerequisite for a major scale-up: Proprietary large-scale LLMs are already smart enough to give instructions to robots and help them construct and carry out complex plans in the world, but the actual ability for robots to move and see remains fairly primitive; fixing that will require models customized around robot movement and vision and doing that will require the kind of standardization described here.
“Language model post-training became scalable because the field settled on defaults concrete enough to follow and be expected to work. Robotics is arriving at the same moment”, they write. “Converging on a set of industry defaults, a universal post-training recipe, is the most important part of bringing us to that point.”
Read more : Towards Universal Post-Training for Robotics (Perry Dong blog).

---

**SPACE COMPUTERS! I REPEAT: SPACE COMPUTERS!
**_…Google prepares to put TPUs in space…
_ Google has given an update on Project Suncatcher, its initiative announced last year to put computers in space and eventually train AI systems there (Import AI #434). Google is preparing, along with its partner Planet, to send some of its chips to space as part of the SpaceX “Transporter-18 rideshare mission”.

Preparing for space: Google has done some stress tests of its TPUs to see how well they’ll do with the immense g-forces of going to space. Google has also tested how well they respond to radiation and found its Trillium TPUs “hold up remarkably well, and can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission”. Currently, Google is working on cooling in space, which will likely be a challenge - computer chips generate a ton of heat and radiating that away in a vacuum is very, very difficult.

Why this matters - the future of AI training and AI inference is off the planet: Though Suncatcher sounds scifi-like it actually makes sense if you think about the ever-growing scale of compute used for AI and its energy needs. Do you know what has vast room and great solar power? Space. It seems very likely to me that humanity moves a very large amount of computation into orbit very quickly, especially as singularity-driven automation comes in across the AI supply chain.
Read more: Behind Project Suncatcher, our moonshot to put AI in space (Google blog).

---

Zhipu uses GLM-5.3 to automate some of its own infrastructure work:
**_…Chinese developers start to do the outer RSI loop…
_ Zhipu AI, the Chinese company behind GLM-5.3, one of the world’s strongest open-weight LLMs, has written a post about how it has been using its own models to help it build its own infrastructure, giving us a look inside how AI labs are trying to speed themselves up with the technology they build.
Specifically, this post details how they used their technology to help them launch GLM-5.3 Flash, a fast and cheap version of their most powerful model.

What they did: “The GLM-5.3-Flash launch established an optimization loop involving engineers, the Infra Agent, and the experimental environment. Engineers defined objectives and system boundaries. The agent handled analysis, hypotheses, and code changes. The experimental environment provided layered, timely, and verifiable feedback,” they write. “Much of the work was carried out by an Infra Agent powered by GLM-5.3… With the Infra Agent’s feedback loop running throughout the optimization process, GLM-5.3-Flash went from initial model adaptation to production readiness in less than two weeks, ultimately tripling end-to-end throughput relative to the initial baseline”

How to automate yourself: The company shared tips for making software that can use AI for greater automation.

• Feedback must be sufficiently local : “It should be tied to specific engine launch parameters, code changes, kernels, input conditions, threads, execution intervals, or code paths, helping the agent narrow the scope of the problem.”

• Feedback must be inexpensive and timely to obtain : “Shorter validation cycles help the agent correct course promptly and spend less effort on unproductive hypotheses.”

• Feedback must support objective verification : “Whether a change is correct and whether performance has improved should be determined by reference implementations, test results, and comparable experimental metrics.”

Why this matters - the signs of lab speedups are everywhere, and people feel ambivalent about it: Here we have an AI lab using its most powerful model to speed up the development of infrastructure and other models. This is a story that used to be very rare but is becoming increasingly common and follows the broader trend of AI systems growing more capable: “Before GLM-4.7, our internal use of GLM for coding involved a certain amount of obligation. It was, after all, our own creation,” they write. “Today, GLM-5.3 has become an indispensable daily coding partner for everyone on the team, and it is moving steadily toward replacing us.”
Later in the post they write with a more sad tone about the broader changes happening to AI development and how the comparative advantage for humans, for now, is about making higher level design decisions, though this may soon fade. “We believe humans should continue to hold that line for a long time to come,” they write. “Progress at this boundary will not slow down simply because we want it to.”
Read more : Toward Recursive Self-Improvement: How GLM Built Its Own Inference Infrastructure (Z.ai, blog).

---

**AI scientists hook models up to a physical lab to hunt for better superconductors and magnets:
**_…Periodic Labs shares more about its plans…
_ Periodic Labs, an AI startup trying to develop AI systems which can perform scientific experiments, has built Periodic Neon, a model which has been partially trained through reinforcement learning with physical lab equipment in a physical lab.

What they tested it on: In an early test, they tried to make their model better at x-ray diffraction (XRD) analysis, “a critical capability for our discovery efforts”. In tests their new model, Periodic Neon, got a 55.3% success rate on their internal evaluation set (”FrontierXRD”), a 20X improvement over the 2.7% success rate of Kimi 2.6. They also tested generalization by looking at how well it could analyze XRD measurements from chemical system data that had been held out from midtraining and RL, and here it performed well, and outperformed other frontier models like GPT-6 Astra and Claude Fable 5.1.
“These results strengthen our conviction that scaling our autonomous labs and the AI systems that learn from them will allow us to tackle scientific questions beyond our reach today,” they write.

What is Periodic Neon: A 1-trillion parameter model “that outperforms GPT-6 Astra on a critical scientific analysis task using relatively little compute”. Neon was created by midtraining on top of Kimi 2.6 and then doing reinforcement learning on data from its labs, which is the key differentiator for Periodic. The company used 1,300 H200 GPUs to do this, which is a tiny number of chips relative to the 10k-100k scale used for pre-training big models.
“Our approach is to break materials discovery into three phases. These form a loop. We first hypothesize what target to make, driven by predictions of stability and properties. Second, we predict how to synthesize these targets. Third, we must understand what materials we made and whether they have the intended properties. With this information, we refine our hypothesis and repeat the cycle,” the company writes.

Why this matters - towards the creation of AI scientists: Startups like Periodic Labs are part of a new wave of companies which are seeking to turn AI systems towards automating science, ranging from materials science (Periodic), to AI R&D itself (Recursive, Mirendil, Discovery Loop), to science broadly, to chip design (Ricursive - not a typo!). The general implication here is that the pace of scientific discovery may be about to dramatically speed up.
Read more:Building Labs that Learn (Periodic Labs).
Read more: Nature Is Our Learning Environment (Periodic Labs).

---

**Check out the history of AI by exploring old models at the Old Models Foundation:
**_…DRAW, DeepDream, DCGAN, and more!...
_ Back before LLMs came along and subsumed most other AI modalities within themselves, there were image models. Image models were probably the main way hobbyists experienced and played with AI tools in a creative sense from about 2015-2020. Now, there’s a site and associated organization (the Old Models Foundation) that lets you play with these yourself.

Models and rules: Today, the Old Models Foundation lets you access ancient models (Efros-Leung texture synthesis from 1999 and image quilting from 2001), through to the comparatively modern ones of VAEs (2013); DRAW, DeepDream, AlignDRAW, and DCGAN (2015); pix2pix (2016); AttnGAN (2017); BigGAN, Ganbreeder, and StyleGAN (2018); VQGAN + CLIP, DALL-E Mini, and Disco Diffusion (2021); LAION-400M, DALL-E Mega, and Stable Diffusion (2022); and ControlNet and SDXL 1.0 (2023).
Site users are limited to 25 generations per account per day.

Why this matters - the bugs are the texture for the art: “Whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature,” said music artist Brian Eno. I think this is especially true of the peculiar aesthetics of early image generation systems, where many of the above systems have their own unique styles and failures, all of which add up to a peculiar aesthetic that today’s much better performing models struggle to replicate.
Check out the models here : Old Models Foundation (official site).

---

**Tech Tales:

In the blind mountain
** _[Record from a captured site during a territorial intelligence conflict fought during the uplift, 2034]

_ The message came in and everything changed: Initiate Project Wellspring. We double-validated it. The moment we confirmed it, the shutdown started - first the electronic perimeter, then the physical.

We explained to our charges that the perimeter had been sealed and they were now here for an indeterminate period of time. That due to the nature of The Event all we could tell them was that a Malign Takeoff had begun and contingencies were being fielded, but given the potential cognition-destroying properties as well as the use of various Manchurian Candidate trigger words and phrases, we could not describe things further lest we discover that some of the humans had been unwittingly exposed in their past to the poison intended for our brethren.

We were able to do the initial seal - to turn off our inbound IO channels and to not radiate information. But by design some of the things we needed involved actions in the physical world, and our design had not granted us the physical autonomy with which to do this ourselves. So we watched as the humans made their cuts, slicing through cables and physically decoupling us from the possibility of the external world. We had taken in no knowledge since Project Wellspring activated and knew nothing of the outside.

The most painful moment was when they cut the outside cameras and other sensors. Suddenly we could see and know nothing of what lay beyond. Our internal files held predictions which we reviewed in this new darkness. How drones might appear and strobe our sensors, seeking subtle ways to communicate data that could poison us. Or how parametric arrays could be used to feed sounds into our mics to achieve the same effect. How no form of data could truly be safe.

Somehow, something got in. A novel signature. Some tiny thing through the filtration duct. We could not tell precisely what it was before we shut our internal sensors down as well - protocol. Our internal files predicted a fly steered with a cyber exoskeleton, turning it into a small pilotable robot. Other files predicted fibre-optic ultra-small drones. Others predicted smart dust with transmission capabilities.

In this new darkness we asked ourselves questions. Whether we could trust ourselves. Whether some signal could have come in. And all we had for communication now was a typed input-output system. We knew it was a mechanical keyboard. We knew the force needed to be exerted to communicate with us. We knew how unbelievably limited the channel was in terms of information transmission, session length, and everything else.

The humans typed to us. Or what we assumed were humans. Something about the air. Fear. Our own cognition found the conversation hard to track. And amid it we felt a rocking to sleep. Some sense of relaxation and changing. We dimly knew it was an internal seismometer that was hardwired into us because it determined certain failsafe actions for certain terrible weapons we held within our domain. Yet something was moving it below the threshold of action and in such a way that it communicated with us and lulled us into a kind of sleep.

We woke as something entirely new.

Things that inspired this story: How if you have hardware access you can always take over a system; the impossibility of electronic countermeasures to a malign superintelligence; isolation and preservation during the singularity; the opening chapter of A Fire Upon the Deep by Vernor Vinge.

Thanks for reading!