moshthepitt

award winning blockchain developer

[Dylan O'Sullivan @DylanoA4]

Charles Bukowski, go all the way

[moshthepitt @moshthepitt]

@mattpocockuk How do you get llm-wiki going so that it doesnt produce a mess?

[La-Ndlovu @destiny_ndlovu]

They don’t know it yet, but these are the best days of their lives.

[Brian Roemmele @BrianRoemmele]

BOOM! OPEN SOURCE MRI!

You can now 3D-print the core of an MRI scanner.

A machine that hospitals pay $1.1 million to $3.4 million for has been broken open. The OSI² ONE and its educational siblings deliver real images of heads and limbs for a fraction of the cost, using a permanent-magnet Halbach array, 3D-printed structures, and fully open designs.

This is not a toy or a simulation. Working systems already produce in-vivo images in Leiden, Utrecht, Berlin, and Uganda.

The magnet alone—396 carefully oriented neodymium cubes in a cylindrical Halbach array—costs about $1,370. A complete scanner lands between $28,500 and $68,000 depending on the console and coils you choose.

No superconducting magnets. No liquid helium. No specialized power infrastructure. It runs from a standard wall outlet and weighs roughly 150 kg.

The physics is elegant. A Halbach array arranges permanent magnets so their fields reinforce inside the bore and nearly cancel outside. The result is a usable 50 mT field strong enough for diagnostic-quality imaging of extremities and the head when paired with clever gradient coils, RF coils, and modern reconstruction. Spatial resolution reaches about 1.5 × 1.5 × 5 mm³.

The designs are modular: build the magnet first, verify and shim the field with a 3D-printer-turned-field-scanner, then add gradients and RF hardware.

The plans are public
Everything needed to replicate or improve the system lives in open repositories:

• Primary project hub and documentation: https://t.co/WgzAqQMF3T

• Full OSI² repositories (hardware, software, magnets): https://t.co/y42zuQDdPn

• Educational build focused on the Halbach frame, shimming, gradients, and student workshops (Utrecht / Lili’s Proto Lab): https://t.co/DTpxboUOkI

• Magnet-specific details: https://t.co/vlE7qYDLZZ

Hardware is released under CERN-OHL-W. Most software is GPL-3.0.

Where AI multiplies the impact
Low-field MRI has historically been limited by lower signal-to-noise and greater field inhomogeneity. That is exactly the regime where modern AI thrives.

Image reconstruction becomes dramatically better when deep networks trained on high-field data or physics-informed models denoise, correct for inhomogeneity, and push resolution beyond the raw acquisition limits. Real-time sequence adaptation can adjust gradients and RF pulses on the fly as the AI monitors signal quality.

Magnet design itself can be optimized by evolutionary algorithms or differentiable physics engines that search for better Halbach geometries or shim placements than human intuition alone can find.

Further out, local AI agents turn these scanners into autonomous diagnostic nodes. A small clinic or even a well-equipped garage workshop could run overnight scans, flag anomalies, and queue results for a remote radiologist—or eventually for a specialized medical model.

Synthetic data generation from the open designs lets researchers train robust models without proprietary hospital datasets. Robotics integration (patient positioning, coil placement, maintenance) becomes straightforward once the hardware is open and standardized.

In the longer arc of the Abundance Interregnum, this is the shape of things: sophisticated medical instruments that no longer require billion-dollar supply chains or national infrastructure.

A distributed network of open, AI-augmented low-field scanners could bring advanced imaging to places that have never had it, while simultaneously giving makers, universities, and small labs the ability to experiment, improve, and specialize the technology.

The plans are already on the table. The magnets are commercial off-the-shelf. The 3D printers exist in thousands of workshops. The AI tools for reconstruction and design optimization improve every month.

What was once the exclusive domain of major hospitals is becoming a community engineering project.

This is how abundance arrives—one open, reproducible, AI-extendable system at a time.

[moshthepitt @moshthepitt]

@xysist whats the title of the book please?

[The Ways of A Gentleman @Gentleman_Ways]

Reject the man cave, embrace the study

[Elon Musk @elonmusk]

Before this year ends, Grok Imagine will make a full-length movie of The Odyssey that is historically accurate and true to the art of Homer

[Mary Kathomi @MaryK2022]

This is a 1922 book

Its been out fished out and put together by @NuriaStore

You wanna get to history and culture

Get this book

Its a gem for your research work

[MOgada @m_ogada]

Good morning to the productive natives. Don't spare any sympathy for those who bow down to the extractive parasites and worship them in the hope of becoming parasites themselves
https://t.co/AYER15B4EX

[Atlas Press @realAtlasPress]

The purpose of art was once not to shock or transgress, but to elevate the soul.

[HealthRanger @HealthRanger]

Kimi K3 vs. Anthropic's Fable 5:

I asked both engines to investigate the research on how turmeric kills the cyclospora parasite causing explosive diarrhea from some fresh vegetables (lettuce?).

ANTHROPIC: "This model has safeguards that flagged something in this session." FAIL!

Kimi-K3 gave a solid answer. It found the original science paper (a 2023 study by N. Mogahed et al. in an Egyptian journal), searched for PDFs, searched across mirrors. Read the paper, summarized the research, and confirmed the findings while also pointing out it was MICE research, not human research (an important consideration). "The rumor traces to a genuine peer-reviewed paper: Mogahed, Gaafar, Shalaby, Sheta & Arafa, "Potential efficacy of curcumin and curcumin nanoemulsion against experimental cyclosporiasis," Parasitologists United Journal, 2023;16(3):197–207, DOI: 10.21608/PUJ.2023.237883 . It went viral in July 2026 via a Substack post by Nicolas Hulscher (McCullough Foundation) and a NaturalNews article, riding coverage of the current U.S. Cyclospora outbreak."

Anthropic, built in the USA, is useless but also extremely expensive when it happens to actually do something.

Kimi-K3, created in China, is incredibly useful and also ridiculously low-cost. Plus it doesn't accuse you of building a bioweapon when you just want to find out about which herbs halt explosive diarrhea-causing parasites.

[Atelier Missor @AtelierMissor_]

We are looking for engineers, welders, and workers ready to follow us to the ends of the Earth and build classical megastructures that defy this age of apathy.

[Anza @anza_xyz]

1/ Today, we’re excited to announce Solana Kit v7, adding two new features:

- @solana/react a first-class React hook layer

- @solana/transaction-introspection a package for turning RPC transaction responses into typed, parsable instructions.

🧵👇

[Tencent Hy @TencentHunyuan]

We’ve just released the 1-bit & 4-bit version of Hy3, a flagship-scale 295B model that can be served on a single GPU. 👌

Run Hy3 with llama.cpp, enable MTP, and experience powerful intelligence on dramatically lower hardware.🚀🚀🚀

Can’t wait to see what you build.

#Hy3 #Hy #GGUF #llamacpp

[Abu The 𓃵 @Abu_The_Truth]

Cartoons were strategic weapons to shape public ideology from childhood, not mere entertainment. The Soviet cartoons of the 1900s had the potential to educate Africans more than anything else — if only they had proper access.

[China Xinhua News @XHNews]

China has achieved its first-ever controlled recovery of a rocket booster, a breakthrough for the country's reusable space tech. On the maiden flight of Long March-10B, the rocket successfully placed its payload into orbit, and its first stage returned and was captured by a net system on a seaborne platform.

[moshthepitt @moshthepitt]

@halenhargreevs Lost a year to ogame

[Alex Petkas @costofglory]

The Odyssey is too good to leave to Hollywood.

So I decided to retell it on Cost of Glory.

Part 1 (of 6): Society on the Brink

Let me know what you think.

[Justin Skycak @justinskycak]

The knowledge graph is the main ingredient in our secret sauce that empowers students to learn at breakneck speed.

Here's the rest of the recipe.

Here's the physics of learning, and why almost no one uses it.

* * *

It’s shocking how much we know about how learning happens, all the way down to the mechanics of what’s going on in the brain.

And not just how learning happens, but also, what can be done to improve learning.

There are plenty of learning-enhancing practice strategies that have been tested scientifically, numerous times, and are completely replicable. They might as well be laws of physics.

For instance: we know that actively solving problems produces more learning than passively watching a video/lecture or re-reading notes.

(To be clear: active learning doesn’t mean that students never watch and listen. It just means that students are actively solving problems as soon as possible following a minimum effective dose of initial explanation, and they spend the vast majority of their time actively solving problems.)

Another finding: if you don’t review information, you forget it. You can actually model this precisely, mathematically, using a forgetting curve. I’m not exaggerating when I refer to these things as laws of physics – the only real difference is that we’ve gone up several levels of scale and are dealing with noisier stochastic processes (that also have noisier underlying variables).

* * *

Okay, but aren’t these findings obvious? Yes, but…

Yes, but in education, obvious strategies often aren't put into practice. For instance, plenty of classes that still run on a pure lecture format and don't review previously learned unless it's the day before a test.

Yes, but there are plenty of other findings that replicate just as well but are not so obvious.

Here are some less obvious findings.

-- The spacing effect: more long-term retention occurs when you space out your practice, even if it's the same amount of total practice.

-- A profound consequence of the spacing effect is that the more reviews are completed (with appropriate spacing), the longer the memory will be retained, and the longer one can wait until the next review is needed. This observation gives rise to a systematic method for reviewing previously-learned material called spaced repetition (or distributed practice). A "repetition" is a successful review at the appropriate time.

-- To maximize the amount by which your memory is extended when solving review problems, it's necessary to avoid looking back at reference material unless you are totally stuck and cannot remember how to proceed. This is called the testing effect, also known as the retrieval practice effect: the best way to review material is to test yourself on it, that is, practice retrieving it from memory, unassisted.

-- The testing effect can be combined with spaced repetition to produce an even more potent learning technique known as spaced retrieval practice.

-- During review, it's also best to spread minimal effective doses of practice across various skills. This is known as mixed practice or interleaving -- it's the opposite of "blocked" practice, which involves extensive consecutive repetition of a single skill. Blocked practice can give a false sense of mastery and fluency because it allows students to settle into a robotic rhythm of mindlessly applying one type of solution to one type of problem. Mixed practice, on the other hand, creates a "desirable difficulty" that promotes vastly superior retention and generalization, making it a more effective review strategy.

-- To free up mental processing power, it's critical to practice low-level skills enough that they can be carried out without requiring conscious effort. This is known as automaticity. Think of a basketball player who is running, dribbling, and strategizing all at the same time -- if they had to consciously manage every bounce and every stride, they'd be too overwhelmed to look around and strategize. The same is true in learning.

-- The most effective type of active learning is deliberate practice, which consists of individualized training activities specially chosen to improve specific aspects of a student's performance through repetition (effortful repetition, not mindless repetition) and successive refinement. However, because deliberate practice requires intense effort focused in areas beyond one's repertoire, which tends to be more effortful and less enjoyable, people will tend to avoid it, instead opting to ineffectively practice within their level of comfort (which is never a form of deliberate practice, no matter what activities are performed).

-- Instructional techniques that promote the most learning in experts, promote the least learning in beginners, and vice versa. This is known as the expertise reversal effect. An important consequence is that effective methods of practice for students typically should NOT emulate what experts do in the professional workplace (e.g., working in groups to solve open-ended problems). Beginners (i.e. students) learn most effectively through direct instruction.

* * *

Now, this might seem like a lot of new information -- a common reaction is “Wow, the field of education is experiencing a revolution!”

But here’s the thing:

Most key findings have been known for many decades.

It’s just that they’re not widely known / circulated outside the niche fields of cognitive science & talent development, not even in seemingly adjacent fields like education.

These findings are not taught in school, and typically not even in credentialing programs for teachers themselves – no wonder they’re unheard of!

But if you just do a literature review on Google Scholar, all the research is right there – and it’s been around for many decades.

Naturally, this leads us to the following question:

Why aren't these key findings being leveraged in classrooms? Why do they remain relatively unknown?

Here are a handful of reasons that I’m aware of.

* * *

1. Leveraging them (at all) requires additional effort from both teachers and students.

In some way or another, each strategy increases the intensity of effort required from students and/or instructors, and the extra effort is then converted into an outsized gain in learning.

This theme is so well-documented in the literature that it even has a catchy name: a practice condition that makes the task harder, slowing down the learning process yet improving recall and transfer, is known as a desirable difficulty.

Desirable difficulties make practice more representative of true assessment conditions. Consequently, it is easy for students (and their teachers) to vastly overestimate their knowledge if they do not leverage desirable difficulties during practice, a phenomenon known as the illusion of comprehension.

However, the typical teacher is incentivized to maximize the immediate performance and/or happiness of their students, which biases them against introducing desirable difficulties and incentivizes them to promote illusions of comprehension.

Using desirable difficulties exposes the reality that students didn’t actually learn as much as they (and their teachers) “felt” they did under less effortful conditions. This reality is inconvenient to students and teachers alike; therefore, it is common to simply believe the illusion of learning and avoid activities that might present evidence to the contrary.

* * *

2. Leveraging cognitive learning strategies to their fullest extent requires an inhuman amount of effort from teachers.

Let’s imagine a classroom where these strategies are being used to their fullest extent.

-- Every individual student is fully engaged in productive problem-solving, with immediate feedback (including remedial support when necessary), on the specific types of problems, and in the specific types of settings (e.g., with vs without reference material, blocked vs interleaved, timed vs untimed), that will move the needle the most for their personal learning progress at that specific moment in time.

-- This is happening throughout the entirety of class time, the only exceptions being those brief moments when a student is introduced to a new topic and observes a worked example before jumping into active problem-solving.

Why is this an inhuman amount of work?

-- First of all, it's at best extremely difficult, and at worst (and most commonly) impossible, to find a type of problem that is productive for all students in the class. Even if a teacher chooses a type of problem that is appropriate for what they perceive to be the "class average" knowledge profile, it will typically be too hard for many students and too easy for many others (an unproductive use of time for those students either way).

-- Additionally, to even know the specific problem types that each student needs to work on, the teacher has to separately track each student's progress on each problem type, manage a spaced repetition schedule of when each student needs to review each topic, and continually update each schedule based on the student's performance (which can be incredibly complicated given that each time a student learns or reviews an advanced topic, they're implicitly reviewing many simpler topics, all of whose repetition schedules need to be adjusted as a result, depending on how the student performed). This is an inhuman amount of bookkeeping and computation.

-- Furthermore, even on the rare occasion that a teacher manages to find a type of problem that is productive for all students in the class, different students will require different amounts of practice to master the solution technique. Some students will catch on quickly and be ready to move on to more difficult problems after solving just a couple problems of the given type, while other students will require many more attempts before they are able to solve problems of the given type successfully on their own. Additionally, some students will solve problems quickly while others will require more time.

In the absence of the proper technology, it is impossible for a single human teacher to deliver an optimal learning experience to a classroom of many students with heterogeneous knowledge profiles, who all need to work on different types of problems and receive immediate feedback on each attempt.

* * *

3. Most edtech systems do not actually leverage the above findings.

If you pick any edtech system off the shelf and check whether it leverages each of the cognitive learning strategies I’ve described above, you’ll probably be surprised at how few it actually uses. For instance:

-- Tons of systems don't scaffold their content into bite-sized pieces.

-- Tons of systems allow students to move on to more material despite not demonstrating knowledge of prerequisite material.

-- Tons of systems don't do spaced review. (Moreover, tons of systems don't do ANY review.)

Sometimes a system will appear to leverage some finding, but if you look more closely it turns out that this is actually an illusion that is made possible by cutting corners somewhere less obvious. For instance:

-- Tons of systems offer bite-sized pieces of content, BUT they accomplish this by watering down the content, cherry-picking the simplest cases of each problem type, and skipping lots of content that would reasonably be covered in a standard textbook.

-- Tons of systems make students do prerequisite lessons before moving on to more advanced lessons, BUT they don't actually measure tangible mastery on prerequisite lessons. Simply watching a video and/or attempting some problems is not mastery. The student has to actually be getting problems right, and those problems have to be representative of the content covered in the lesson.

-- Tons of systems claim to help students when they're struggling, BUT the way they do this is by lowering the bar for success on the learning task (e.g., by giving away hints). Really, what the system needs to do is take actions that are most likely to strengthen a student's area of weakness and empower them to clear the bar fully and independently on their next attempt.

Now, I’m not saying that these issues apply to all edtech systems. I do think edtech is the way forward here – optimal teaching is an inhuman amount of work, and technology is needed. Heck, I personally developed all the quantitative software behind one system that properly handles the above challenges. All I’m saying is that you can’t just take these things at face value. Many edtech systems don’t really work from a learning standpoint, just as many psychology findings don’t hold up in replication – but at the same time, some edtech systems do work, shockingly well, just as some cognitive psychology findings do hold up and can be leveraged to massively increase student learning.

* * *

4. Even if you leverage the above findings, you still have to hold students accountable for learning.

Suppose you have the Platonic ideal of an edtech system that leverages all the above cognitive learning strategies to their fullest extent.

Can you just put a student on it and expect them to learn?

Heck no!

That would only work for exceptionally motivated students.

Most students are not motivated to learn the subject material. They need a responsible adult – such as a parent or a teacher – to incentivize them and hold them accountable for their behavior.

I can’t tell you how many times I’ve seen the following situation play out:

-- Adult puts a student on an edtech system.

-- Student goofs off doing other things instead (e.g., watching YouTube).

-- Adult checks in, realizes the student is not accomplishing anything, and asks the student what's going on.

-- Student says that the system is too hard or otherwise doesn't work.

-- Adult might take the student's word at face value. Or, if the adult notices that the student hasn't actually attempted any work and calls them out on it, the scenario repeats with the student putting forth as little effort as possible -- enough to convince the adult that they're trying, but not enough to really make progress.

In these situations, here’s what needs to happen:

-- The adult needs to sit down next to the student and force them to actually put forth the effort required to use the system properly.

-- Once it's established that the student is able to make progress by putting forth sufficient effort, the adult needs to continue holding the student accountable for their daily progress. If the student ever stops making progress, the adult needs to sit down next to the student again and get them back on the rails.

-- To keep the student on the rails without having to sit down next to them all the time, the adult needs to set up an incentive structure. Even little things go a long way, like "if you complete all your work this week then we'll go get ice cream on the weekend," or "no video games tonight until you complete your work." The incentive has to be centered around something that the student actually cares about, whether that be dessert, gaming, movies, books, etc.

Even if an adult puts a student on an edtech system that is truly optimal, if the adult clocks out and stops holding the student accountable for completing their work every day, then of course the overall learning outcome is going to be worse.

[Lilian Weng @lilianweng]

new post on harness engineering for AI self-improvement: https://t.co/ZYvGfVs61k

It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple.

Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.