Entrepreneur dedicated to decentralisation & jurisdictional arbitrage. Advocate of technological innovation & sovereignty. Sometimes help out at animal shelters

Birmingham
Joined June 2022
C. Hill retweeted
traffic from ChatGPT is skyrocketing on my apps and Ilias shares a gem here this is such a good framework side note: i looked into it and blog posts are 73% of what AI looks at when replying blogging is faaarr from being dead
Want ChatGPT to recommend your SaaS? 1. Ask AI: "best [your category] tools" 2. Copy competitors it recommends 3. Paste to Ahrefs → Backlinks 4. Filter: "Best links" + Sort by "Page Traffic" 5. Find "Best X" listicles linking to them Email them, you'll show up in ChatGPT!
87
18
6
368
40,584
C. Hill retweeted
I want to give a shoutout to @coderabbitai for letting me do free code reviews on open source github repos. Claude codes well, but coderabbit always finds the edge cases.
1,049,817 open-source code reviews in September, a new monthly record! That's 2.49x April's volume and up 14.4% from August. "Not using it today would be like racing Formula One without a seatbelt or helmet." – @tannerlinsley, creator of TanStack
2
5
12
586
C. Hill retweeted
Building an AI workplace means stitching together APIs, agents, and real-time messaging. This repo gives you the server core. OCTO Server is the Go backend for OCTO, an open workplace for humans and AI agents. It helps you build the backend layer for an agent-enabled collaboration product by centralizing REST and WebSocket APIs, Lobster agent scheduling, and the WuKongIM messaging control plane. Key features: • REST + WebSocket APIs – handles conversations, users, groups, files, organizations, and webhooks • Agent orchestration – routes Lobster sessions and tool-call execution inside the server • Real-time messaging control – drives WuKongIM for channels, messages, and presence • Pluggable data services – includes MySQL-compatible migrations, cache repositories, and object-storage adapters • Deployable Go service – build the binary with Go and start it from the included YAML configuration template It’s open-source (Apache License 2.0). Link in the reply 👇
6
10
30
1,826
Co-founder of @OxideComputer Bryan Cantrill (@bcantrill) spoke with @TBPN about venture capital's relationship with untraditional bets. His observation: investors say they want contrarian ideas, but few commit when the bet could plausibly return nothing. "Everybody loves contrarian. But no one actually wants a zero." Watch the full clip below:
Oxide co-founder @bcantrill believes every VC wants to be contrarian, but most get cold feet when an actual contrarian bet is staring them in the face. "Everybody loves contrarian. But no one actually wants a zero."
4
1
8
346
C. Hill retweeted
// Latent Semantic Memory for LLMs // This paper presents some really interesting findings about agent memory. They find that sixteen memory tokens beat full-context inference on FactKG while using 108x fewer tokens for the source documents. Approach: Documents are encoded once into reusable embeddings. For each query, a learned module gathers the relevant information from the whole collection and writes it into 16 tokens in the input space of a frozen LLM. Results: It beats Mem0, MemoRAG, and other baselines on five of eight benchmarks. On FactKG, it scores 84.33% against 80.43% with the full context. It also works as agent memory. In TextWorld, using the agent's own history as the collection raises success from 18.6% to 24.2%, and the agent revisits fewer rooms. It also seems that it could help frozen LLMs accumulate and use knowledge effectively across long-context interactions. This is applicable to building continually learning AI systems. arxiv.org/abs/2610.21375 Chat with Paper: academy.dair.ai/papers/memor…
55
36
3
213
15,660
C. Hill retweeted
Personal agents get harder when they need to use real apps and respect real boundaries. Open Instinct is an open-source personal-agent repository for builders who want to run an agent they can text. It helps you move from a message to an agent-run task by combining a text-facing identity, an agent loop in a Linux desktop microVM, connected apps, and policy checks before tools run. Key features: • Text-based identity – Inkbox supplies a phone number, email address, and agent-to-agent endpoint • Isolated computer – the agent runs in its own microVM with a Linux desktop for real task execution • Trust tiers and grants – seven code-enforced access levels and plain-English exceptions govern tool use • Connected apps and editable behavior – Composio integrations and file-based prompts let you connect apps and adjust the agent • Payment approval flow – Stripe Link creates a one-time card for an exact total, with owner approval on the phone It’s open-source (MIT license). Link in the reply 👇
4
11
2
30
2,193
British firms using AI report a net 8% job loss, twice the international average. US firms with similar productivity gains are hiring. Same technology. Employers pay £3,750 in National Insurance on a £30,000 job, and nothing on the AI that replaces it.
1
11
1
32
3,467
C. Hill retweeted
Capital gains tax is theft, committed twice against the same dollar. You earn income. The government takes its cut immediately, in the US its between 22% and 37% depending on your bracket. What remains is yours, after-tax money, already processed through Washington's collection machine. You choose not to spend it. You save it, invest it, buy assets with it. The discipline required to delay consumption is precisely the behavior that funds capital formation across the economy. Then the asset grows. You sell it. The government appears again, demanding 20% (plus the 3.8% Net Investment Income Tax for higher earners) on the nominal gain, which often includes pure inflation adjustments that represent no real increase in purchasing power whatsoever. The IRS treats your patience as a second taxable event. The original dollar was already taxed. The "gain" frequently reflects nothing more than the dollar losing value while you held the asset. Murray Rothbard identified this pattern: the state punishes capital accumulation. Consider a simple case. You invest $50,000 of after-tax earnings in 2016 into an S&P 500 index fund. By 2026, it's worth $115,000. Inflation consumed a meaningful slice of that $65,000 gain. You pay capital gains tax on the full nominal figure regardless. Capital gains taxation suppresses investment, shrinks the capital pool available for productive enterprise, and forces entrepreneurs to seek financing from a shallower pond. Fewer productive investments get made. Workers carry that cost through lower wages and fewer opportunities. The state expands, while you suffer.
48
209
22
835
12,065
. In just five years, TSMC and Samsung clearly left China's Alibaba and Tencent behind In 2020, Chinese tech companies were still leading Asia’s market cap rankings. - Alibaba was worth about $625 billion and Tencent around $700 billion. - Samsung was near $500 billion, while TSMC stood at roughly $490 billion. Five years later, the picture completely changed. - TSMC surged past $2 trillion, becoming Asia’s most valuable company. - Samsung also jumped strongly, reaching over $1.1 trillion. - Meanwhile, Alibaba fell to around $270 billion and Tencent dropped below $500 billion. The shift was driven by the AI boom: TSMC produces the world’s most advanced chips, and Samsung dominates high-end memory chips used in AI systems. Chinese internet companies, by contrast, faced regulatory pressure and weaker growth. As a result, Taiwan and South Korea’s semiconductor leaders now clearly outrank China’s biggest tech firms in market value. Source: CompaniesMarketCap
8
15
1
57
2,655
Replying to @DavidSteinrueck
.@jensei_ reco to talk about ourvPrivacy Market Report with @DavidSteinrueck — Aztec can be a good supporter fit.
1
1
1
93
Claude Code just got an official plugin directory that hit 41k stars in days Anthropic is now maintaining a public marketplace of high-quality plugins so you can extend Claude Code without hunting random repos github.com/anthropics/claude…
31
28
2
317
67,384
C. Hill retweeted
Thames Water asks creditors for another £2.4bn. Up to 30% of TW revenue covers financial costs. Debt spiral deepens problems, it borrowed to pay dividends. Hedge funds funding TW, hoping to asset-strip, hike bills, bailout. Let it go bust. Nationalise. theguardian.com/business/202…
38
430
10
615
11,205
Some tech companies use algorithms to amplify the negative because it is more engaging and viral and drives higher ad revenue - even though it reduces the well-being of individuals and society. But a former journalist told me yesterday that hyper-local content — something that happened near where you live — gets higher engagement than negative content!! This is music to my ears as my company seeks to launch 10,263 local daily news sites in partnership with local editors and journalists. So promising!
3
1
1
7
438
TSLA, Samsung: second Taylor fab breaks ground in November, 2nm production targeted 2030 + Samsung breaks ground on Taylor Fab 2 this November, a 2.7M sq ft building one and a half times bigger than Fab 1, with mass production targeted for 2030 on 2nm. Fab 1 reaches full operations by end of 2026 with volume production in 2027, its best known customer Tesla with a $16.5B deal running through 2033. + Samsung plans over $37B of investment across Central Texas, backed by up to $4.745B in CHIPS Act funding. Samsung is putting over $37B into Central Texas while TSMC's Q3 profit jumped 68.7% to NT$763B on AI driven demand, per ainvest's Oct 15 coverage. US 2nm class capacity is now being built by two foundries at once.
3
4
36
5,602
This fully local AI job tool just crossed 49k stars and runs entirely on Claude Code It evaluates job postings, rewrites your CV, and preps interviews without sending anything to the cloud If you’re struggling with applying to jobs you need this github.com/MadsLorentzen/ai-…
20
13
1
130
58,507
It seems SA was too bullish on the new GW added per year in the next few years?
13
7
7
125
47,436
New: European VCs I spoke with report a lack of transparency from some startups around revenue, as well as confusion around the different types of revenue — in some cases to an extreme.  I dug into the "blurry" world of defence revenue for @Siftedeu sifted.eu/articles/blurry-wo…
1
8
372
C. Hill retweeted
Your agent fleet shouldn’t lose the job when one process dies. JarvisCore is a Python framework for builders creating AI agent systems, from a single agent through distributed multi-agent workflows. It helps you run agents across a peer-to-peer mesh by handling capability discovery, shared workflow state, and recovery beneath the agents you write. Key features: • Peer-to-peer mesh – agents discover one another by capability over SWIM gossip, without a central orchestrator • Durable workflow state – Redis-backed state can recover completed workflow steps when you rerun the same workflow ID • Zero-trust credentials – Nexus injects scoped, encrypted credentials into integrations so agents do not handle raw keys • Typed integration atoms – add a Python function as an atom, validate it with the CLI, and register it for runtime discovery • Built-in observability – inspect per-step traces and use the bundled Prometheus and Grafana setup for metrics It’s open-source (Apache License 2.0). Link in the reply 👇
5
9
26
1,843
Our next generation of open weights multimodal decision models comes in two sizes, pplx-decider-v1.2 and pplx-decider-flash-v1.2. Text and vision scores improve. Inference is up to 5x faster: all questions in a request run in one pass. Flash is the top 4B on the Decision Index.
28
25
1
217
32,992