Les Yetton (@LesYetton) header
Les Yetton (@LesYetton) profile picture

Curious AI lab rat | Stage 1 Ventures | CEO | Investor | Board member | Advisor | Promoter of all things boating

Boston, MA
Joined February 2012
Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)Photo by Les Yetton (@LesYetton)
Exchange soft honest helpful feedback with trusted close pals gentle kind thoughtful suggestions support mutual growth maintain clear transparent reliable long term friendly bonds alive
1
50
Les Yetton retweeted
Static post-training is not enough. The industry is hooked on the illusion that one-off fine-tuning or prompt-stuffing solves enterprise execution. It doesn't. Context rots and static models break. The real frontier isn't post-training, it's continuous learning: building runtime systems where human judgment continuously compounds into durable, in-model capability. That is why Jensen’s ecosystem strategy is brilliant. Hugging Face is huge for American leadership, but the Poolside transaction may matter even more for compound intelligence. Jensen will bring American open-weight AI to the frontier, likely with a multi-billion dollar training run for a 10T+ parameter Nemotron model in the next 18 months. The game-changer? Bundling Nemotron base models directly with LatentSpin’s smart agent solution. Knowledge workers can build, teach, train, and deploy secure vertical AI across the enterprise, continuously adapting in real time at a fraction of human operational costs. When the best open-weight foundation is American, cheap to run, and built for continuous learning, organizations finally own their intelligence instead of renting it from walled gardens. Diversity of open AI safeguards human freedom, and cheap specialized models will only make frontier reasoning tokens more valuable. Stop renting static intelligence. Start compounding it. #LatentSpin #ContinuousLearning #Nvidia #Nemotron #OpenWeightAI #AgenticAI #EnterpriseAI #SovereignAI #CompoundingIntelligence
1
3
35
Les Yetton retweeted
Judgment-rich environments are where AI actually moves the needle, and Healthcare Revenue Cycle Management (RCM) is sitting right at the absolute center of that intersection. Healthcare RCM isn't just about moving data from point A to point B, it requires continuous interpretation, complex clinical context, and navigating an endless maze of payer-specific nuances where human expertise has historically been the only safety net. That’s why LatentSpin is such a natural fit here. By moving beyond brittle automation into architectures built for complex continuous learning and nuanced reasoning, we're helping teams turn RCM back into a strategic advantage instead of a operational bottleneck. We’re currently starting our ecosystem and actively looking for innovative design partners in the healthcare space to build the future of intelligent workflows together. If you're tackling RCM complexity head-on, let’s talk. #HealthcareAI #RCM #HealthTech #AI #LatentSpin #EnterpriseAI
1
2
54
Open AI poked the bear, and now Jensen Huang is going all-in to compete with them. It all blew up after OpenAI announced "Jalapeno", their own custom inference chips that directly threaten Nvidia's business model. NVIDIA's response? Pulling the plug on investments and speed-running open source by acquiring Hugging Face and Poolside, plus dropping a full open-source AI stack next year. Six months ago, people thought open source couldn't touch OpenAI. Now, the world's most valuable chip company owns the premier open-source AI platform. OpenAI just created his ultimate death threat. #AI #Nvidia #OpenAI #TechNews #JensenHuang
1
3
169
NVIDIA is becoming the Central Bank of AI.
3
94
This is interesting.. Salesforce and Anthropic launched Claudeforce, a Claude Cowork plugin with 37 pre-built sales skills that lets sellers query, update and act on live CRM data without opening Salesforce, part of a stock-boosting earnings beat. Is Salesforce taking a back seat? venturebeat.com/orchestratio…
3
161
Nvidia spent 12 days of revenue ($12.9B) to buy Hugging Face. Why? To protect the open-weight ecosystem keeping developers locked into CUDA while major labs build custom silicon escape routes. For new platform builders like @LatentSpin and others on the open source train, owning distribution and open model access isn't just strategic, it’s the foundation for real continuous learning architectures. #AI #Nvidia #HuggingFace #OpenSource #OpenWeight #CUDA #EnterpriseAI
1
69
This is wild... OpenAI just dropped chip benchmarks beating Nvidia's flagship at half the power (700W vs 1,400W), just one week after Nvidia backstopped $105B of OpenAI's data center debt. The breakdown on public benchmarks: ➡️Efficiency: 1.5x–1.9x more throughput per kW than Nvidia GB200/GB300 racks ➡️Speed: Up to 3.6x lower latency ➡️Versatility: Beached Nvidia on DeepSeek R1 and Kimi K2.5 too—this is general inference, not a one-off ASIC The part breaking everyone's brain is the timeline: 16 months from first hire to ready to roll. Industry norm is 3-4 years. This directly speaks to the power of AI. OpenAI used its own AI models to design the chip that will run its models. The self-improving feedback loop is officially shipping silicon. Nvidia is now in the wild-west position of financing the exact data centers where its biggest customer is displacing its hardware. Rollout starts later this year. Link in comments..
2
2
2
183
Just read where Goldman Sachs expects AI infrastructure spending to surge from $765B in 2026 to $1.64T by 2031.
1
3
141
Les Yetton retweeted
Big step forward from @databricks on agentic extraction. Their move to decompose tasks into parallel agentic workflows validates a core belief at @LatentSpin: for complex enterprise tasks, the "model" is not the solution, the orchestration harness is. 🤖⚖️ But high-accuracy extraction is only the first mile. The real opportunity is turning that extracted data into a proprietary asset. At LatentSpin, we’re taking the output of these robust extraction harnesses and feeding them into a continuous learning architecture. We don't just want to extract the data; we want to capture the domain expertise of the humans correcting it, so the agent improves in real-time. While others focus on the infrastructure to "get the data out," we’re focused on the feedback loop that makes your agents smarter with every single document they process. The future isn't just "precision" extraction... it's sovereign, self-improving intelligence. 🚀 #EnterpriseAI #AgenticAI #LatentSpin #ContinuousLearning #SovereignAI Databricks blog link in comments...
1
1
2
80
Les Yetton retweeted
Beyond Graph Engineering: From Human Intent to Strategy Synthesis 🚀 First there was Prompt Engineering. Then Context Engineering. Now Graph Engineering. Each makes agents more capable, but humans still have to engineer every part of the solution. What's next? Strategy Synthesis. At Latentspin, founder @ThomasHazel believes the future is telling the system what you want to accomplish, not how to accomplish it. Let the system synthesize the strategy, build an explicit plan, learn what it needs, and execute reliably. Read more: lnkd.in/gFrPujye #Latentspin #AI #GraphEngineering #StrategySynthesis #ArtificialIntelligence #TechTrends
2
1
48
So Claude is watermarked. When do they come out with a "for fee" non watermarked version?
1
3
141
BSP, where SaaS goes to thrive, up 12% today... Go figure...
1
2
80
Read this today and it makes sense. Building agents is like someone giving you 1,000 legos and saying go build the Empire State Building... You have no idea how to get started.. But if someone gives you the Empire State Building kit with a set of instructions and videos to follow, different story.. There needs to be more plug this here and plug that there kits out there.
193
I'm surprised we haven't seen SPAC-ish type newly formed PE going after the SaaS rollup play. There has to be a ton of 1+1+1 = 10.
Credit to @bendingspoons for having the 🏀 🏀 to pull off these deals when no one else will This rundown is a bit blind to actual structure, but a fine place to start I’m guessing there would be interest for those last-stage investors, but as @bgurley pointed out, a lot of late-stage SaaS was done as straight money — chasing a lot of challenging dynamics on the board level Here we are 5 years past peak SaaS, and we are finally sorting out these deals — which is a good sign Congrats to @airtable on a fantastic product and run… shocked Google, MSFT, and other major players didn’t jump on here There’s so much SaaS to buy right now, but I guess it’s too small compared to AI for folks to find an internal champion I would love to fund a roll-up vehicle of the smaller SaaS startups; < $100m in revenue
1
251
Les Yetton retweeted
The open-weights train has no brakes. 🚂💥 Alibaba’s Qwen team has consistently been setting the standard for open SOTA, from coding and math benchmarks to inference efficiency and long-context performance. Every major drop they make reshapes the baseline for what developers can build locally and in production. Super excited to see what they’re dropping next. If history is any indicator, we’re about to get another massive leap in capability for the open-source AI ecosystem. Eyes locked on this one. What are you hoping to see most... better reasoning, agentic coding capabilities, or multimodal performance? 👇 #Qwen #OpenSourceAI #LLMs #MachineLearning #AI #LatentSpace #OpenWeights
📢Meet Qwen3.8-Max — our most capable model to date. Next week, the open weights of Qwen3.8-Max will be released, and Qwen3.8-27B is also going open-weights to meet you all!🎉 Qwen3.8-Max, a new bar for coding and cowork at 2.4T parameters: - Autonomous coding: 10+ days of self-evolving development, from empty folder to production without hand-holding, complete project trace in the GitHub:github.com/qwen-code-dev-bot… - Real work, real results: Production-quality deliverables across hundreds of professions. - Long-horizon mastery: System-level autonomous planning with closed-loop adaptive learning, driving 500+ turns of chip design optimization and 365 days of e-commerce strategy. - Native multimodal intelligence: Vision isn't just input — it's a continuous feedback loop for planning, execution, and self-correction. 💰Pricing: Input: $2.0 / M tokens Output: $6.0 / M tokens Implicit Caching: $0.25 / M tokens Start building with Qwen3.8-Max! 🚀 📖 Blog: qwen.ai/blog?id=qwen3.8 ✅ Qwen Studio: chat.qwen.ai/?models=qwen3.8… ⚡ API: qwencloud.com/models/qwen3.8…
1
43
Bending Spoons buys Airtable for $1.28B... Not a good time to be a legacy SaaS business. $1.4B raised $11B valuation $480M rev growing 20% YoY  $1.285B enterprise value $2.25B equity value  $965M of net cash on its balance sheet.
1
1
81
Same prompt, different cost. Like getting a burger at Capital Grille or a food truck. They both taste great..
1
46