Ekram hossain

Research Data Engineer II @ URMC Medical Center| Data Scientist @ ROCHCI lab | Applied Machine Learning | Human-centered (Situated) Machine Intelligence | UX/UI

[Codez @0xCodez]

https://t.co/eJB3TgyJwV

[Akshay 🚀 @akshay_pachaar]

CPU vs GPU vs TPU vs NPU vs LPU, explained visually:

5 hardware architectures power AI today.

Each one makes a fundamentally different tradeoff between flexibility, parallelism, and memory access.

> CPU

It is built for general-purpose computing. A few powerful cores handle complex logic, branching, and system-level tasks.

It has deep cache hierarchies and off-chip main memory (DRAM). It's great for operating systems, databases, and decision-heavy code, but not that great for repetitive math like matrix multiplications.

> GPU

Instead of a few powerful cores, GPUs spread work across thousands of smaller cores that all execute the same instruction on different data.

This is why GPUs dominate AI training. The parallelism maps directly to the kind of math neural networks need.

> TPU

They go one step further with specialization.

The core compute unit is a grid of multiply-accumulate (MAC) units where data flows through in a wave pattern.

Weights enter from one side, activations from the other, and partial results propagate without going back to memory each time.

The entire execution is compiler-controlled, not hardware-scheduled. Google designed TPUs specifically for neural network workloads.

> NPU

This is an edge-optimized variant.

The architecture is built around a Neural Compute Engine packed with MAC arrays and on-chip SRAM, but instead of high-bandwidth memory (HBM), NPUs use low-power system memory.

The design goal is to run inference at single-digit watt power budgets, like smartphones, wearables, and IoT devices.

Apple Neural Engine and Intel's NPU follow this pattern.

> LPU (Language Processing Unit)

This is the newest entrant, by Groq.

The architecture removes off-chip memory from the critical path entirely. All weight storage lives in on-chip SRAM.

Execution is fully deterministic and compiler-scheduled, which means zero cache misses and zero runtime scheduling overhead.

The tradeoff is that it provides limited memory per chip, which means you need hundreds of chips linked together to serve a single large model. But the latency advantage is real.

AI compute has evolved from general-purpose flexibility (CPU) to extreme specialization (LPU). Each step trades some level of generality for efficiency.

The visual below maps the internal architecture of all five side by side.

👉 Over to you: Which of these 5 have you actually worked with or deployed on?

[Rony @Ronycoder]

Instead of watching an hour of Netflix, watch this 30-minute speech by the Head of Anthropic’s Coding Agents research team. It will teach you more about vibe coding than 100 paid courses.

[Akshay 🚀 @akshay_pachaar]

A harnessed LLM agent.

Most people picture this as a model with tools bolted on. The real architecture inverts that relationship.

The model itself is deliberately thin. Intelligence gets pushed outward, and the harness composes it at runtime.

Three dimensions orbit the harness core:

𝗠𝗲𝗺𝗼𝗿𝘆 holds state the model shouldn't carry in weights or context. Working context, semantic knowledge, episodic experience, and personalized memory each have their own lifecycle.

𝗦𝗸𝗶𝗹𝗹𝘀 hold procedural knowledge. Operational procedures, decision heuristics, and normative constraints specialize the general model per task.

𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹𝘀 hold the interaction contracts. Agent-to-user, agent-to-agent, and agent-to-tools are three distinct surfaces with their own failure modes.

Between the core and these modules sit the mediators: sandboxing, observability, compression, evaluation, approval loops, and sub-agent orchestration. They govern how the harness reaches out and how state flows back in.

The useful question this framing unlocks: for any new capability, where should it live? Stable knowledge goes to memory, learned playbooks go to skills, communication contracts go to protocols, loop governance goes to the mediators.

Harness design becomes a question of what to externalize, and how to mediate it.

I'm building a minimal agent harness from scratch. Didactic, easy to read, no magic. Open-sourcing it soon. Stay tuned.

[Garry Tan @garrytan]

If your memory dies when your harness dies, you built the harness too thick.

Memory is markdown. Skills are markdown. Brain is a git repo. The harness is a thin conductor — it reads the files, it doesn't own them.

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[Andrej Karpathy @karpathy]

- Drafted a blog post
- Used an LLM to meticulously improve the argument over 4 hours.
- Wow, feeling great, it’s so convincing!
- Fun idea let’s ask it to argue the opposite.
- LLM demolishes the entire argument and convinces me that the opposite is in fact true.
- lol

The LLMs may elicit an opinion when asked but are extremely competent in arguing almost any direction. This is actually super useful as a tool for forming your own opinions, just make sure to ask different directions and be careful with the sycophancy.

[Brian Roemmele @BrianRoemmele]

Astonishing work by @varun_mathur!

We are testing this now!

[ericosiu @ericosiu]

https://t.co/LGxhaM4TDn

[Meer | AI Tools & News @Meer_AIIT]

https://t.co/huTGfXkL3S

[Y Combinator @ycombinator]

Congrats to @LemonSliceAI on their $10.5M seed!

They built the world's first interactive talking AI video model—a face layer for voice agents. Trained on a custom, 20B-parameter video diffusion transformer, streaming at 20fps on a single GPU. Infinite-length video generation with no error accumulation.

https://t.co/jyHm3lZqNZ

[Luke Auburn @lukeauburn]

Could Alexa help people find out if they have Parkinson's disease? Computer scientists at @UofR developed a speech-based AI tool that can screen for a key indicator of the neurodegenerative disease #URochesterResearch Story:https://t.co/pV5IiAd0dZ
Study: https://t.co/rm4VzXDlhc

[Ehsan Hoque @ehsan_hoque]

I am deeply humbled to have been awarded the Presidential Early Career Award for Scientists and Engineers (PECASE) by President Biden. This is the highest recognition by the U.S. government to scientists and engineers.

There's an interesting paradox in academia where students put in tremendous effort, yet professors get all the credit. This honor truly belongs to my students and research collaborators, who have been both teachers and learners alongside me throughout our journey together.

My late mother, who passed away at 48, taught me the straightforward formula for success - maintain your presence and never surrender. This moment would have meant so much to her.

@UofR @UofRDataSci

https://t.co/MovFLiXc2i

[maharshi @maharshii]

super nice blog by @ezyang on pytorch internals

[Yoshua Bengio @Yoshua_Bengio]

@HopfieldJohn and @geoffreyhinton, along with collaborators, have created a beautiful and insightful bridge between physics and AI. They invented neural networks that were not only inspired by the brain, but also by central notions in physics such as energy, temperature, system dynamics, energy barriers, the role of randomness and noise, connecting the local properties, e.g., of atoms or neurons, to global ones like entropy and attractors. And they went beyond the physics to show how these ideas could give rise to memory, learning and generative models; concepts which are still at the forefront of modern AI research.

Their ideas inspired me so profoundly that I decided to choose learning in neural networks for my own research as a graduate student. They motivated me to look for abstract principles that could be as simple as the laws of physics, but could explain biological as well as artificial intelligence. I'm truly delighted for them and for our field.

[Pinaki Bhattacharya @PinakiTweetsBD]

A video of Sheikh Hasina's Home Minister, Asaduzzaman, has gone viral. In the video, a police officer is seen telling Asaduzzaman, "When we shoot, one person dies and another gets injured. Only one goes down, sir, the rest don't. This is, sir, the biggest terror."

A secretary is also seen in the video. This secretary, Jahangir Alam, is currently the secretary to the Home Advisor Sakhawat Hossain. This is the first clear, public admission of shooting directly to kill. It happened with the knowledge and approval of the top leadership. There can be no greater evidence of a crime against humanity than this.

[Rumor Scanner @RumorScanner]

X (formerly Twitter) is flooded with Bangladesh related communal rumors.

In the last 4 days, Rumor Scanner team has identified at least 20 such rumors spread by Indian accounts.

In the first part of these communal rumors, a total of 10 fact-checks are provided below...

[Jeremy Howard @jeremyphoward]

For those that hope (or worry) that LLMs will do breakthrough scientific research, I've got good (or bad) news:

LLMs are particularly, exceedingly, marvellously ill-suited to this task. (if you're a researcher, you'll have noticed this already)

Here's why🧵

[Google AI @GoogleAI]

Announcing DataPerf, a set of new #ML challenges that ask participants to measure and validate data-centric algorithms and techniques to create and improve datasets using various benchmarks. Learn more and sign up → https://t.co/TDm1KWhngx