Viral

Viral AI News (Oct 3): EBAI, ds4 & Muse Gadgets

Viral AI news for Oct 3, 2026: EBAI's parody site mocks Washington's AI rebrand, antirez runs frontier models locally with ds4, and Muse Gadgets takes off.

Daniyal

· 4 min read
Editorial illustration of a parody corporate badge beside a laptop running a local AI model

The most viral AI story of the last 24 hours was a joke. EBAI, a parody consultancy with a deadpan website, climbed to the top of Hacker News as a satire of Washington's decision to rename artificial intelligence as "super intelligence." Behind it: the Redis creator shipping a way to run frontier open weights on a Mac, Aleph Alpha releasing a sovereign German model, and Meta's gadget platform hitting the front page on day one. Yesterday's edition covered Pi 1.0, DeepSeek and the 217 year cipher, and the viral archive has the rest.

EBAI: the parody site that beat the policy it mocks

EBAI is a fake consulting brand whose landing page borrows the language of federal AI policy, complete with service lines for sovereign intelligence and a deadpan corporate voice. It reached the top of Hacker News on Oct 3 with 389 points and 85 comments. The joke lands because the target is real: on Sept 29 President Trump signed an executive order titled Inaugurating The Era Of Super Intelligence, which directs executive branch agencies to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI." Source: The White House Source: EBAI

The joke is the gap between a naming directive and an actual capability. Satire, not achievement.

ds4: frontier open weights running on your own laptop

DwarfStar 4, or ds4, is a narrow inference engine written in C by Salvatore Sanfilippo, the creator of Redis, and released under the MIT license. It runs large open weight models locally on high memory Macs, CUDA boxes and ROCm machines, including DeepSeek V4 and V4.1 Flash, GLM 5.x and Qwen3.8 Flash Next. The trick is asymmetric 2-bit quantization aimed at the routed experts of a mixture-of-experts model while keeping the critical shared paths precise, plus a KV cache that spills to SSD so a restart does not force a full re-prefill. The Hacker News thread reached 296 points and 89 comments. Source: DwarfStar 4 Source: GitHub

The numbers are the interesting part. On an M5 Max with 128GB the project reports 790 tokens per second of prefill and 39.4 tokens per second of generation at a 2,048 token context, dropping to 398.5 and 27.6 at 65,536 tokens. On a DGX Spark with 128GB it reports 825.8 prefill and 18.1 generation. That is a 284 billion parameter mixture-of-experts model answering from local memory.

Kolibri: a 78B sovereign model that beats models four times its active size

Kolibri is an open weight language model from Aleph Alpha, released on Oct 3 under Apache 2.0. It is an English and German mixture-of-experts with 78 billion total parameters and 3 billion active per token, with a context window of up to one million tokens. It scored 96.9 on AIME 2025, 84.3 on GPQA Diamond and 85.9 on LiveCodeBench v6, and Aleph Alpha claims it matches models with up to four times its active parameter count. The company trained it in Germany and Finland and says 21.3% of pretraining tokens were organic German text. Source: Aleph Alpha Source: Hugging Face

Reach was split across two submissions: Aleph Alpha's own post drew 135 points and 23 comments, while a third party write up drew 218 points and 107 comments. The larger second number means people read the analysis, not the press release.

Muse Gadgets: Meta open sources hardware for its AI agent

Muse Gadgets is a platform for building small devices around Muse, Meta's personal AI agent. On Oct 2 the company open sourced the gadget SDK, published a token system and project ideas, and the resulting page drew 220 points and 98 comments on Hacker News. The projects are deliberately small: ESP32 boards that take prompts and return answers from a desktop assistant. Source: The Verge Source: Muse Gadgets Source: GitHub

Hacker News comments were clear about what it is: an SDK for tinkering, not a new class of hardware. It is still a distribution play, because every gadget built on it embeds Meta's agent somewhere new.

Quick Comparison

Product or Trend What Went Viral Engagement Metric Verdict
EBAI Parody of the super intelligence rename 389 HN points, 85 comments Satire, but the target is real
ds4 Frontier model running locally in C 296 HN points, 89 comments Substance, verified benchmarks
Kolibri 78B sovereign open weight model 353 HN points across two posts Substance, strong math scores
Muse Gadgets Open sourced AI agent hardware SDK 220 HN points, 98 comments Distribution play, small devices

Frequently Asked Questions

What is the viral AI news today?

EBAI, a parody consultancy mocking the federal rebrand of AI as super intelligence, is the most viral AI story today. It hit the top of Hacker News on Oct 3 with 389 points and 85 comments. The joke targets a real Sept 29 executive order directing agencies to use "Super Intelligence" instead of "Artificial Intelligence."

Can you really run frontier models locally with ds4?

Yes, and the reported numbers are usable for chat and coding on high memory machines. DwarfStar 4, written in C by Redis creator Salvatore Sanfilippo, reports 39.4 tokens per second of generation on an M5 Max with 128GB at a 2,048 token context, falling to 27.6 at 65,536 tokens.

What is Kolibri and why did it go viral?

Kolibri is Aleph Alpha's 78 billion parameter open weight model for German and English, released Oct 3 under Apache 2.0. It activates only 3 billion parameters per token and scored 96.9 on AIME 2025 and 84.3 on GPQA Diamond. Two separate Hacker News posts about it drew 353 points combined.

Background worth having before you act on the above:

Building an AI agent, automation, or backend workflow? AsaanKaroobar builds custom AI solutions, apps, and digital experiences. Get in touch →

More on Viral

Browse every Viral story →