![Flux[dev]: A cavernous engineering library made of concrete. Seated at a desk reading is a futuristic robot, sleek smooth humanoid design. Smooth, glossy black faceplate with no visible facial features, high-tech, minimalist appearance. The robot's body is matte black or dark gray, with articulated joints and mechanical parts that resemble those of a human, including fingers. The word "Tech" is carved into the concrete wall in large letters.](https://ethanbholland.com/wp-content/uploads/2024/09/tech-2.png)
Flux[dev]: A cavernous engineering library made of concrete. Seated at a desk reading is a futuristic robot, sleek smooth humanoid design. Smooth, glossy black faceplate with no visible facial features, high-tech, minimalist appearance. The robot’s body is matte black or dark gray, with articulated joints and mechanical parts that resemble those of a human, including fingers. The word “Tech” is carved into the concrete wall in large letters.
Kalmogorov-Arnold Neural Networks Shake Up How AI Is Done – IEEE Spectrum
“📌 Fantastic paper from @AIatMeta – substantial FLOPs savings while maintaining or improving performance with modality-aware mixture-of-experts (MoE) architecture designed for pre-training mixed-modal, early-fusion language models. MoMa introduces modality-specific expert
“The paper explores how similarity-based in-context example selection impacts low-resource machine translation (MT) using large language models (LLMs). It finds that similarity-based selection significantly improves translation quality for low-resource languages, unlike previous
“LG AI Research presents EXAONE-3.0 model:
“Model merging has become popular recently due to its ability to easily combine the capabilities of multiple LLMs. Here’s how it works, why it’s useful, and a few (practical) examples… What is model merging? Model merging is a technique that operates in the parameter space of a
How GitHub harnesses AI to transform customer feedback into action – The GitHub Blog
[2311.17035] Scalable Extraction of Training Data from (Production) Language Models
“Really cool new FastHTML project from: a growing gallery of live FastHTML code examples. See the next tweet to learn how you can contribute your own sweet components, templates, and applications. Try it here:
“Interconnect bandwidth and latency is crucial for these workloads, so you really can’t distribute like that, but some numbers for fun: Xbox One S: 1.4 TF Xbox One X: 6 TF H100: 1000 to 4000 TF So, 100k H100 > all 30M current generation Xbox.” / X
“What are we doing with this capital? Originally we intended to raise $300M which was going to allow us to deploy 108,000 LPUs into production by end of Q1 2025. We raised 2x that, so we’re also expanding our cloud and core engineering teams. We’re hiring!” / X
“this is an extremely interesting way to judge an llm what do you think @repligate @simonw @DanielMiessler @hrishioa” / X
“We are humbled to announce the close of Thrive IX. Exceeding $5 billion, Thrive IX comprises $1 billion designated for early-stage investments and $4 billion designated for growth-stage investments. At Thrive we aim to be concentrated in both people and ideas. The ways in which
My benchmark for large language models
“Ron Jeffries apologized for inventing Story Points. He then went to call the whole idea of software estimation “evil.” Story points is the Emperor’s New Clothes. Everyone I’ve seen—and I mean everyone—uses points as a measure of time: 1 point = 2 hours 3 points = 1 day 5
“Direct Preference Optimization (DPO) has become one of the go-to methods to align large language models (LLMs) more closely with user preferences. If you want to learn how it works, I coded it from scratch:
“MLX best practice: use lazy loading to reduce peak memory use.
“Procedural & GenAI Combining the control and flexibility of procedural design in Processing with generative AI opens up new creative possibilities, enabling us to explore complex geometric solutions with instant feedback. Geometric constraints and rules can easily be established
Chat with Open Large Language Models
“Base 3 Computing Beats Binary
How I Use “AI”
AI Tech Stack https://twitter.com/jowyang/status/1821949136129667574
Leave a Reply