Keuxdi

🆕 esper release!

Google Gemma 270M: a tiny blazing fast model. Fun to play with, surprisingly good for our smallest supported model.

Liquid AI LFM2-VL: A tiny, fast and efficient vision model designed to run on phones.

Numerous quality of life improvements and bug fixes!

I think I fixed sycophancy

I think I fixed sycophancy

🧠 Qwen3 2507 thinking and instruct are now available in esper 💻 gpt-oss 20b is also available if you want to use esper on your M series mac ✨ Added Google Gemini to providers making it easier to setup your API key to access models, control reasoning with GPT-5, and much more

🧠 Qwen3 2507 thinking and instruct are now available in esper

💻 gpt-oss 20b is also available if you want to use esper on your M series mac

✨ Added Google Gemini to providers making it easier to setup your API key to access models, control reasoning with GPT-5, and much more

pivoting

pivoting

We're all doomed

We're all doomed

SoftBank and OpenAI’s $500 Billion AI Project Struggles to Get Off Ground - WSJ

SoftBank and OpenAI’s $500 Billion AI Project Struggles to Get Off Ground - WSJ

Grok 4 alignment team

Grok 4 alignment team

Honestly my nightmare

Honestly my nightmare

The entire Meta superintelligence team in 4 years after collecting their cheques

The entire Meta superintelligence team in 4 years after collecting their cheques

This was clickbait. It's $1B now.

This was clickbait. It's $1B now.

3 new models now available in esper! @LiquidAI_'s LFM2 is designed specifically as an on-device model. It compares well to much larger models while being much faster and memory efficient. I've also added Falcon H1 🇦🇪 and LG's hybrid reasoner EXAONE 🇰🇷 This brings the total to over 20 mo

3 new models now available in esper!

@LiquidAI_'s LFM2 is designed specifically as an on-device model. It compares well to much larger models while being much faster and memory efficient.

I've also added Falcon H1 🇦🇪 and LG's hybrid reasoner EXAONE 🇰🇷

This brings the total to over 20 mo

"Meta’s superintelligence lab considers shift to closed AI model" - NYT

"Meta’s superintelligence lab considers shift to closed AI model" - NYT

When it’s my turn during standup and I get to show off the button I made this week

When it’s my turn during standup and I get to show off the button I made this week

Meta’s new super intelligence team wants to create a new closed model

Meta’s new super intelligence team wants to create a new closed model

I ❤️ local inference

I ❤️ local inference

Openai ~4B open model Grok 2/3 open sourced Meta superintelligence lab edge model

Openai ~4B open model

Grok 2/3 open sourced

Meta superintelligence lab edge model

The model weights are on huggingface

The model weights are on huggingface

Local model summer Everybody’s talking about it

Local model summer

Everybody’s talking about it

*AGI being pursued by the least ethical companies and people* Me:

*AGI being pursued by the least ethical companies and people*

Me:

Prepping for Grok 4 and new OpenAI model release

Prepping for Grok 4 and new OpenAI model release

Grok alignment team

Grok alignment team

On the eve of Grok 4, trained on the biggest cluster ever, we get a SOTA mobile model from Hugging Face. Open weights, open data, open methods. Huge respect to the team for sharing everything.

On the eve of Grok 4, trained on the biggest cluster ever, we get a SOTA mobile model from Hugging Face.

Open weights, open data, open methods. Huge respect to the team for sharing everything.

🤔

🤔

Where are the open Grok 2 weights you promised? Elon:

Where are the open Grok 2 weights you promised?

Elon:

Me: I literally have LLMs running on my phone Everyone:

Me: I literally have LLMs running on my phone

Everyone:

The first model under 10B with LMArena score of 1300+. Running on a phone.

The first model under 10B with LMArena score of 1300+. Running on a phone.

In 2007, it was unfathomable to have a powerful computer in your pocket. Today, it is unfathomable to have superintelligence in your pocket.

In 2007, it was unfathomable to have a powerful computer in your pocket.

Today, it is unfathomable to have superintelligence in your pocket.

Using Gemma 3n on my phone while waiting for openAI's open-weights model

Using Gemma 3n on my phone while waiting for openAI's open-weights model

What kind of "open" AI are you?

What kind of "open" AI are you?

o3 not sharing its reasoning tokens

o3 not sharing its reasoning tokens