Ogiste

better access, knowledge x opportunity 🌍🇰🇪
creator economy + guerrilla philosopher
founder building @sasasasa_labs

contributoor
@TheSafariDAO @UTUtrust

[Ogiste @og1ste]

Gm gm 🫡 it's been a minute, so here's a short recap. Spent the past 10 years working, learning and building consumer software. I spent the last four years working, learning and web3 tech x ecosystems. Joined a few hackathons trying to do too much 😂

[Don Pettit @astro_Pettit]

I finally have all 1.2 million raw image files from my latest mission to ISS! Here is a sample of one of my favorite Milky Way photos, taken from the Cupola with Nikon Z9, Arri Zeiss 15mm lens, T1.8 with custom sidereal drive that cancelled out star motion relative to our orbit.

[Ogiste @og1ste]

Uncover the secrets of the universe

[Ogiste @og1ste]

What is the ultimate use of frontier technology?

[Ogiste @og1ste]

feed the kids, heal the sick

[calle 🟥 @callebtc]

India wants to ban Bitchat.

GitHub asked us to delete the code.

We won't.

Leave a star ⭐️ if you agree 👇

[jack @jack]

https://t.co/WFWSz5jEpc

[Naval @naval]

When software was expensive - thin, horizontal, best-of-breed software stacks extracted rents across every business.

Now that software is cheap - value moves to vertically integrated businesses that deliver opinionated end-to-end experiences.

[jack @jack]

we're launching BUZZ!

a new groupchat platform for teams of people and agents of all sizes, built to reduce our dependency on slack and github. model-agnostic, decentralized, self-sovereign, and open source. 🐝

https://t.co/8IaMVeTQNo

[Mark A100 @gatere_mark]

I’m excited to share our new paper:
From a Multilingual Streaming ASR Backbone to Kenyan-Language Systems: Data-Centric Adaptation of Nemotron 3.5 for Kikuyu, Dholuo, and Kalenjin

The phrase “low resource” is often interpreted as “too little audio,” but building real-world ASR systems reveals that the challenge is much broader.

A language may have hundreds of hours of recordings and still be difficult to model because transcriptions contain inconsistent spelling, non-speech events appear as lexical tokens, audio files are missing, duration tails destabilize batching, or dataset splits are not as independent as their names suggest.

At @c_elolabs, we have been working through these challenges end to end, from raw speech data to deployed, true-streaming automatic speech recognition.

The paper documents how we adapted @NVIDIAAI Nemotron 3.5 ASR using @PyTorch and NeMo, including:

• Corpus auditing and language-specific normalization
• Full-parameter continuation training
• Cache-aware true-streaming evaluation
• Checkpoint selection, preservation, and deployment
• Failures and data-quality decisions that shaped the final systems

Our internal tests during model development showed:

Kikuyu ASR
WER: 42.97%
No-space CER: 7.79%

Dholuo ASR
WER: 33.98%
Conventional CER: 9.59%
No-space CER: 8.13%

Both models support live, incremental speech transcription through isolated language-specific streaming services.

Kalenjin remains a work in progress. The paper transparently documents its data-cleaning journey, evaluation limitations, and the requirements for a valid confirmatory protocol.

Read the paper: https://t.co/dvnK3QB2jv

I welcome feedback from African-language ASR researchers, linguists, and practitioners.

#AfricanAI #AutomaticSpeechRecognition

[Cheng Lou @_chenglou]

As the quantity of code explodes in the next few years, I hope the folks worrying about quality can rest assured that there’s evidence that our ability to verify even the most casual vibe code can keep up with the amount of slop

Turns out that math was the ultimate spec

[Arena.ai @arena]

Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.

This is a 17-place jump from Kimi-k2.6 (#18 -> #1).

In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.

The full model weights will be released by July 27.

Congrats to the @Kimi_Moonshot team on this major milestone!

[Stephen Deng @mrstephendeng]

Here's some quick thoughts on K3 as it relates to AI builders in frontier markets like Africa.

1. First, K3 weakens the distillation-only story around Chinese open-sourced models. The gap has fundamentally shrunk. Model margins are going to crush and converge everywhere which wasn't always the assumed path.

2. Assuming open-source models will remain open (K3 weights release later this month) and have closed the gap, the capability floor on local fine-tuning rises with the tide. This is critical for regions like Africa where the edge-cases abound and reliability is paramount. Africa will need to generate its own verification loops.

3. This deepens the use case for local data center buildouts, something I was hesitant about before. While I still don't think the continent is going to produce competitive general models, the ROI for local inference also rises as fine-tuned models permeate.

4. As for on-prem, there's a ton of memes about how it's impossible for individuals to run K3 at home, but these are all short-sighted when we include frontier market enterprise/gov. We are talking about <$10m in server costs (and falling) and 6-figures of energy per year for on-prem, frontier-level intelligence. Incredible.

5. This massively increases the surface area for market-tuned models, especially for sovereignty-sensitive verticals that will demand on-shore models, inference, and data retention.

6. The bottlenecks move to talent, partnerships, and capital. IMO, all are solvable.

[vitalik.eth @VitalikButerin]

A ten-thousand word monster post trying to cover the entire tech tree behind the main lineage of obfuscation (iO) protocols:

https://t.co/46nseINlwF

Special thanks to all who helped!

[Ogiste @og1ste]

The industrial era turned heat into action.
The agentic era can turn imagination into action.

[Mission 44 @mission44]

We never forgot! 🐐👏

[Lewis Hamilton @LewisHamilton]

REMEMBER WHO YOU ARE

[Nataliecrue.eth @NatalieCrue]

S/o to the homie @bigvibessss for bringing back Dat Piff and doing the challenging work of archiving Hip Hop history

[Nataliecrue.eth @NatalieCrue]

S/o to the homie @bigvibessss for bringing back Dat Piff and doing the challenging work of archiving Hip Hop history

[Eren Chen @ErenChenAI]

Mocap actress @xixiyusakana Real Time Teleoperating 6 robots in an AMD sponsored event.

[Ogiste @og1ste]

Shukran to @ArtporiumKe having me. The team was asking all the right questions and guiding thoughtful discussions on culture. Cant wait for the next one. Still asking people around me what does art mean to you?

[Ogiste @og1ste]

Come to the 254 Beer District this afternoon for good music and great convos. I'll be representing @sasasasa_labs and speaking with other thoughtful culture leaders on music, tech and mindset 🌱