Hardware for Deep learning

GPU

 

Context engineering

 context engineering GUI

space it is

 less than $20/kg

Is space investable?

orbit raising

Interchip connect

 tpu system-architecture-tpu-vm

AWS Gravitron

WIndsurf

 Building-windsurf-with-varun-mohan

Riptide LLM

top leetcode patterns coding interview questions

cline persistent memory through rules

Claude session ID , git and restore

cursor best practices

developer productivity

pay the cost of coupling or decoupling 

tidy first

Information leakage, deep vs shallow classes, layers should remove complexity

bunch of LLMs

AMD beyond-code-autocomplete

Scaling Intelligence: Designing Data Centers for Next-Gen Language Models

Andrej Karpathy

 Andrej Karpathy JN talk Software 3.0 and building with agents - Software is changing

Other talks/podcasts I enjoyed lately

cursor team with Lex Fridman Future of Programming

Windsurf Varun Mohan

Claude and Cursor

 Claude making own commands

organizing AI project

cursor rules

cursor prompt

A Pattern Language for agentic AI

Context programming

 Context programming

Building with cursor and others

 circuit breaker

rulesfile - can be used with Windsurf too

create and update a logbook

PBI - Product Backlog items

PRD - Production rules?

Anatomy of Cursor

multiple models -  embedding models

                               char models

                               diff apply models

Autonomy slider mental model

Anatomy of Perplexity

Claude in Japanese

AI verifying

FIFO questions

Anatomy of an AI assistant

 With varying answers by the AI assistant even with the same underlying models, we have to come up with a max of approach. If multiple AI assistants agree on the approach, then go ahead with the change.

We probably need another wrapper for this to happen.

With time, each will start offering what the other does -- going towards homogenity so that they can provide all under one umbrella or stick to their niches if that is their strength -- on the consumer/enterprise scale. Just the enterprise too has a vast range of simple to complex environments. This makes one wonder, how the scale problem has been solved by the ones that handle complex codebase.

The ones that go with a tool that handles their simpler situation now, will have to plan for the transition to the one that can handle their code size growth in future.

Alpines ceo-in-training program

 Alpines ceo-in-training program 

Building Alpine

deeper technology

"The hard thing that we've built is connecting to 80,000 stores across the U.S., connecting to 1,100 retailers and this ability to translate your intent of cooking something for dinner into products showing up at your doors within 2 hours because the physical world is still going to remain pretty difficult to navigate even in an AI-enabled world."

This is from chatgpt-----

[User Intent Input]

       |

       v

+------------------+

|  Intent Parser   | <-- GPT-4 / LLM

| ("I want to cook |

|  pasta tonight") |

+------------------+

       |

       v

+------------------+

| Recipe Resolver  | <-- Maps to canonical ingredients

| (e.g., chicken,  |

| garlic, pasta)   |

+------------------+

       |

       v

+------------------------+

| SKU Mapper & Matcher  | <-- Vector DB (e.g., Pinecone)

| Ingredient → Localized|

| products from nearby  |

| stores                |

+------------------------+

       |

       v

+----------------------------+

| Inventory + Price Checker | <-- Real-time APIs or data feeds

+----------------------------+

       |

       v

+----------------------------+

| Store Selection Engine     | <-- ETA, stock, proximity

+----------------------------+

       |

       v

+-------------------------+

| Delivery Orchestrator  |

| (dispatch, tracking)   |

+-------------------------+

       |

       v

+-------------------------+

| Frontend (Mobile/Web)  |

| User sees cart, selects|

| subs, tracks delivery  |

+-------------------------+

The journey of a prompt

 From prompt to the next in Cline

RISC V

 sirinsoftware inside risc-v microarchitecture

risc vs increasing influence

Memory for AI

 Data pre-processing by CPU+LPDDR

Learning/inference with GPU+HBM - paurooteri

Implementing AI activation functions ie nonlinear functions for inference, shows why it comes at the cost of performance.

The best DRAMS for AI

Algorithms for Inference MIT course

Compared to alternative solutions, its packet-based processing reduces external memory moves by more than 75% for Llama 3.2 1B and Qwen2 1.5 B. Even in highly memory-bound use cases, Origin Evolution excels in producing 1000s of effective TFLOPS and dozens of tokens per second per mm2 of silicon.

"Running LLM inference in edge hardware is crucial because it reduces latency and eliminates security concerns associated with cloud-based implementations. However, deploying LLMs in resource-constrained systems poses challenges due to their large model sizes and significant computational requirements. Consequently, edge designs require specialized hardware that can effectively address their unique resource constraints, including power, performance, area (PPA), latency, and memory requirements. Moreover, innovative software optimizations are essential, including model compression, hardware optimization, attention optimization, and the creation of dedicated frameworks to manage computational and energy constraints at the edge."

ARCHITECT

 Architecture, form, space and order

101 Things I Learned in Architecture

AI beginners

organizing AI code 

boss like

I saw a kids t-shirt and read the first part

～のような ボス 

~ No yōna bosu

a boss like

I guessed from the picture of a dinosaur that ボ might be bo(ss). Feels good to be able to recognise a language I am learning and understand it.