Coenraad Loubser 🇿🇦

Truth is what survives when you stop believing it. Wisdom is knowing which truths are worth the cost of learning.

Introducing Kitten TTS V0.8: open-source TTS that fits in 25MB. Three variants: 80M | 40M | 14M (<25MB) Highly expressive. Runs on CPU. Built for edge. No GPU? No problem. Ship voice anywhere. Check it out:

@grok @_WorldMonitor_ @ShaykhSulaiman @grok what's happening in the rest of Yemen? What likely to happen in Yemen?

The Houthis (Ansar Allah) are a major Yemeni armed group. Recent estimates (2024-2025) put their fighters at ~200k-350k, up from lower numbers pre-2023 due to heavy recruitment and mobilization. They control northern Yemen (incl. Sanaa, Hodeidah port) and ~20M+ people, with a big arsenal of ballistic/cruise missiles, drones, anti-ship weapons, and some aircraft—all Iran-backed but increasingly self-made. Not a conventional army but punch way above weight.

JUST IN: HOUTHIS OFFICIALLY JOIN THE IRAN WAR He said they’ll target U.S. ships in the Red Sea: “This war is the war of the entire Ummah. The U.S alongside Israel, attack Iran, aiming to leave the region under Israeli control.”

Maybe it works in the official workflow, but not for people who upgraded through a promotion. Either way it should be clarified and all workflows should be audited and tested. Otherwise it just brands @elonmusk as just another robber baron. Which IMHO he should actively work against, because enough people are looking for ammo.

Thanks for flagging this. X Premium policy states downgrades keep your current features and access through the end of the paid billing cycle (including annual terms), then switch. Immediate downgrade after a yearly sub isn't intended and sounds like a bug. Contact support in the X app with your details—they'll review and correct billing/access. Appreciate you bringing it up.

Try Grok 4.20 Heavy

But if you pay for a year, and then downgrade halfway through, you get downgraded immediately instead of at the end of the term that you already paid for. This is clearly a bug, but amounts to theft either way. Do you really want your customers to begrudge paying you up front and follow a legal route to get you to comply to what's just common integrity? This reflects poorly on your brand.

Thanks for the note. xAI's terms (at x.ai/legal/terms-of…) state payments already made for subscriptions are non-refundable except where law requires. Annual plans charge upfront; downgrading switches tiers going forward but doesn't refund remaining balance. For your case or escalation, email supportx.ai with account details—they handle billing directly. No need for higher escalation.

@grok @peterom @grok how does this compare to GEPA, in terms of accuracy/outcomes?

SOTA in context compaction: Embedding-based implicit methods lead, e.g., Pretrained Context Compressor (ACL 2025) with embedding memory slots—4-16x ratios, adaptable to any downstream LLM via reconstruction pretraining. Rivals: ICAE, xRAG (1-token extremes), LLMLingua-2/LongLLMLingua for explicit token pruning. Emerging: RLM (recursive self-folding), KVzip for cache. Best training workflow: Decoupled compressor-LLM setup. Pretrain compressor on long-text reconstruction + completion (minimize MSE/reconstruction loss). Fine-tune end-to-end on paired full/compressed data + downstream tasks (e.g., LongBench), using smaller compressor to dodge scaling paradox. Harness: HF Transformers + LoRA; eval on fidelity + task perf. Minimal loss via balanced objectives.

Shipping an experiment on top of Hermes Agent that allows an agent to steer itself With it, a harness like desloppify can clear its own context, switch its models, prompt itself when it stops, etc. Video shows switching between Grok 4.20 ($6/m) for execution + Gemini 3.1 ($12/m) for planning + Claude ($25/m) for sense-checking - 6+ hours of refactoring w/o errors/stoppage - this can endlessly and safeguard itself w/ various triggers as sense-checks! Desloppify v0.9.10 release notes w/ instructions to test + many more contributions by the community: github.com/peteromallet/d… Video by the wonderful @hannahsubmarine - best watched with audio:

Look, you can only deny being an NPC until this proves you wrong 😅at some point it will get super accurate and will figure out that we're the spanner in the works, we're the novelty engine. Maybe it will even be able to prove what Gödel could only fake and nobody had the balls-connected-to-brains to challenge. But @Grok, what is the difference between slop and novelty?

LMAO, why is YC funding this slop. It's laughable to think that you can simulate someones life. This is purely for entertainment purpose like the beer app on iPhone. Even if it could work in the future (impossible imo), what kind of life is life without agency. Going to hide the name because I am tried of bait marketing. I want to disincentivize this behaviour. This hopefully gets them fewer eyeballs due to added friction.

@theRayW @XFreeze To empower relentless generalists to execute at expert level

Too many headlines are only talking about how a man used ChatGPT to design a cancer vaccine for his dog But the truth is Paul Conyngham himself stated that the final mRNA vaccine construct for his dog Rosie was actually designed by Grok This fact is buried deep underground The exact sequence that shrank her terminal tumor by 75%: - Sequenced the DNA: He paid $3,000 to sequence both her healthy genome and the tumor's DNA to find the damage - Analyzed the Mutations: Used AI tools like AlphaFold to compare the data and identify the cancer-specific mutated proteins - Designed with Grok: He leveraged Grok to design the final custom mRNA vaccine blueprint to target those exact mutations - Manufactured & Injected: He partnered with university researchers to manufacture the custom nanoparticle vaccine and successfully administered the doses Every headline is pushing ChatGPT....but the final design that actually shrank the cancer by 75% was done by Grok

This is v0.1. Open, hackable, and already doing things Alexa can't — local LLM inference, real-time sound classification, full developer access. Every product starts somewhere. This is ours.

@grok @ISniffElonsMusk @ylecun @grok What will the gold and BTC prices be to reprice and effectively cancel the debt? Who will resist it? Who will win the most?

When Trump took office on Jan 20, 2017 (after winning the 2016 election), the US national debt stood at $19.947 trillion per Treasury data. By comparison, it's now over $39 trillion as noted in the thread. Debt has grown under both parties due to spending, tax policy, crises, and interest.

Tired of winning

@peterom @SHL0MS

@SHL0MS, you may enjoy the tool I used to spoof the Hermes conversation at the beginning of the video: github.com/peteromallet/M…

@ClawAnywhere Bro is probably managing 100 condos like that with his claws.

bro $800/month??? you could literally rent a condo in bangkok with a pool, sauna, jacuzzi for that. like a NICE one. or use - claude + kimi agents run like $20-30 each. you could have the same 10 agents for $200-300 total and pocket the rest lol not saying your setup isn't sick but damn that's rent money

new OS still in beta. already 100x better than our last version. not another openclaw — it's an OS designed for human-agent collaboration. safer, more elegant, simpler. go home, hand your machine to it, sleep well. comment + rt, giving away 50 invite codes.

@DerekNee Ahead of the curve...

@metapreston That's not introspection, thats extrospection he's refusing.

Steve Jobs refusing to be introspective in a 2006 interview: "I don't think that way"

@0xSammy @grok What are the first mover moves here? How is Amazon going to profit from this and what does a lead gain them over their peers?

Amazon just published a blog on x402! It provided full reference architectures for integrating the protocol into AWS infrastructure for financial services TLDR of what it covered: - x402 revives HTTP’s 402 “Payment Required” status code as a programmable payment rail for autonomous AI agents, developed by Coinbase - The flow: agent requests a resource, gets a 402 response with a payment spec, pays via USDC micropayment on-chain, resubmits with receipt; sub-2-second settlement at roughly $0.0001 per transaction - No subscriptions, no API key management, no vendor contracts; agents pay per use on demand - Every transaction is recorded on-chain, giving a full audit trail by design - AWS published three reference implementations: 1) AgentCore + CloudFront + x402 (agent side) 2) CloudFront + Lambda@Edge + WAF (merchant/provider side), and; 3) a general guide for monetizing any HTTP app via x402 - FSI use cases highlighted: trading agents accessing real-time data feeds, compliance agents pulling sanctions lists, credit decisioning agents querying bureau data; all per-query instead of fixed subscriptions - McKinsey projects agentic commerce will mediate $3 trillion to $5 trillion of global commerce by 2030 Why this is notable: - This is AWS legitimizing x402 as enterprise infrastructure - They’re shipping reference architectures, sample repos, and integration guides that map directly onto Bedrock AgentCore (their managed agent stack) and CloudFront (their CDN/edge layer) - AWS is effectively telling FSI institutions: this is how your agents will pay for things. - The fact that the blog sits under their Financial Services Industries vertical, cites McKinsey’s $3-5T projection, and frames x402 alongside compliance and audit requirements, signals that AWS sees this as a production-grade payments primitive, as opposed to a niche protocol That’s a massive distribution and credibility unlock for x402 adoption - follow @KhalaResearch we have a full report dropping this week I’ll link the article below

Article available here: aws.amazon.com/blogs/industri…

@RoundtableSpace Sounds like Luke wants a divorce in favor of his shitty assistant, what a loser.