SF Compute launches low-cost automated AI training with fair queues

SF Compute rolled out an automated research product that lets users train models at low cost, with fair queues, no long-term contracts, and MCP-based ticket recovery for agent-driven training jobs. — via 1
Alex Lieberman says almost no one is broadly using AI agents yet, and for most people AI still equals ChatGPT; the industry is still very early and mass agent usage would strain compute. — via 1 2
acquire.com listed several cash-flowing micro-SaaS assets, including an AI vector graphics platform at $99K and a GHL Marketplace app at $180K, confirming steady demand for small digital businesses. — via 1 2
Postiz grew from $17K to $171K MRR in two years, with influencer shoutouts as a measurable customer-acquisition channel. — via 1 2
levelsio's self-tracking data links personal-training days to +20% productivity, while travel and sickness are the biggest productivity killers. — via 1 2

SF Compute released an automated research product that lets users run model training at low cost. It uses a fair queue so a few users can't monopolize resources, avoids long-term contracts, and includes a built-in MCP ticket mechanism that lets agents automatically resume training tasks after platform failures. This lowers the barrier for indie developers and small teams to do serious model work. — via 1
Ben Tossell trained a 30B model to play the board game 1830, spending about $2–3K. The model didn't learn new financial concepts, but it significantly reduced hallucinated facts and improved citation faithfulness; validation with GoodfireAI's tools showed higher precision/recall and better planning. This is a practical example of cheap, focused model experimentation. — via 1

Alex Lieberman argues that almost no one is truly using AI agents at scale; for most people, AI still means ChatGPT-style chat. The biggest mainstream experience shift is ChatGPT being able to send emails and create calendar invites, which shows how early the industry is. If billions of people ever rely on agents for real work, compute will come under enormous pressure. — via 1 2
Levelsio warns against letting AI rewrite a large codebase in one massive automated 'Gauntlet Loop' pass, saying it produces messy, inefficient code and burns ~$500; he prefers controlling AI step-by-step per feature/object, calling the viral approach a gimmick that's not mature. — via 1

acquire.com's new listings illustrate an active market for small digital assets: an AI vector graphics platform with $26.7K LTM revenue, 24% MRR growth in two months, 70K accounts, and Google ranking #1–2 is asking $99K; the top social app SaaS on GHL Marketplace (with $6.2K MRR and 4.4% monthly churn) is asking $180K; an AI stock trading app with $11.2K LTM revenue is asking $30K; and an Amazon KDP research/publishing platform with $16.8K LTM revenue is asking $59.5K. The spread shows serious buyers still hunting for small, cash-flowing niches. — via 1 2 3 4
Flippa highlights a profitable, cash-flow-positive business with zero SEO as an excellent acquisition target; a buyer found a $150K hidden gem that everyone overlooked. This reinforces that under-marketed businesses can be attractive deals for buyers who do manual screening. — via 1
Starter Story's Postiz case study shows a SaaS scaling from $17K to $143K MRR, and the founder now has two hires (support and development). He pays influencers $60–350 per shoutout for Postiz quotes, and notes that this isn't cheap when normalized for scale. Two years ago Postiz didn't exist, and MRR was $0; it's now at $171K MRR. — via 1 2

Nick Huber says his portfolio companies employ 400 people, 375 of them outside the U.S. (Colombia, South Africa, Philippines), with monthly revenue over $5M. He defends outsourcing, arguing that critics who buy Chinese-made goods while condemning offshore labor ignore that the U.S. has neither enough workers nor the same quality of life without it. — via 1 2
On real estate, Huber notes that bigger players are moving to larger deals, leaving room for small transactions; he focuses on $500K–$1M projects. Two early deals in Pittsburgh ($475K) and Erie ($650K) generate about $130K in annual passive cash flow while the assets appreciate. — via 1
Levelsio's ongoing self-tracking shows personal-training days with high intensity correlate with +20% productivity, even after including training and shower time, while working from co-working spaces correlates with −7% productivity (though he still values it socially). The worst productivity hits are flying and getting sick. He cautions these are directional hypotheses, not statistically significant results. — via 1 2
Patrick McKenzie used an LLM to organize ten years of angel-investing documents, bank statements, and transfer exports, calling it a win for LLM-powered personal data wrangling that avoided manual spreadsheet work. — via 1

Discover