Mason Ba

AI Startup, Founding Member | Materials Science, Engineering @ Carnegie Mellon | Product & Project Dev

[Mason Ba @PhaseNO]

Qwen has released Qwen-AgentWorld, an open-source world model (Apache 2.0 license) designed to simulate agent environments across multiple domains.
Ideal for advancing reliable, scalable AI agents.
Details: https://t.co/6tNMHLhbkY
#AI #OpenSource #LLM #Agents

[Mason Ba @PhaseNO]

IBM has unveiled the world's first sub-1 nanometer chip technology, featuring a revolutionary 0.7 nm (7 angstrom) "NanoStack" 3D architecture.
https://t.co/iQ5B7cpWWx

[Mason Ba @PhaseNO]

No major new frontier model releases, but agentic AI momentum is building fast.
arXiv: https://t.co/cDosY7iiRR Papers: https://t.co/eoee37nCR0
#AI #AgenticAI #MachineLearning

[Mason Ba @PhaseNO]

Key new papers include:

OpenThoughts-Agent: Data recipes for stronger agents
MemGUI-Agent & MobileForge: Long-horizon mobile GUI agents
Qwen-AgentWorld: Language world models for general agents. 👇🏻👇🏻👇🏻

[Mason Ba @PhaseNO]

OpenAI has launched 'Patch the Planet,' a new Daybreak initiative in partnership with Trail of Bits to help open-source maintainers identify, validate, and fix critical vulnerabilities using AI-assisted security research and expert human review.

[Mason Ba @PhaseNO]

Strong arXiv activity in https://t.co/5Ib74ORCzS, featuring new papers on MoE calibration under distribution shift, LedgerAgent for policy-compliant tool-calling, Multi-LCB for multilingual code benchmarks, and FP4 pretraining optimizations.
Full papers: https://t.co/b1oPDKCN8f

[Mason Ba @PhaseNO]

https://t.co/FKTV6jKctG releases GLM-5.2: A new open-weight frontier model!

753B MoE, stable 1M context, excels on SWE-bench & agentic coding. MIT license, ready for local deployment.

Open-source AI continues to close the gap. Developer win!

#AI #OpenSource #LLM #GLM52

[Mason Ba @PhaseNO]

For startups, the real opportunity is not just building on top of models, but connecting them to real-world workflows, proprietary data, rigorous evals, and deep domain knowledge.

#AIStartups #Agents #AGI #AIEvals #DomainKnowledge #EnterpriseAI

[Mason Ba @PhaseNO]

The next leap toward AGI won’t come from bigger language models alone, but from systems that can understand the world, take long-horizon actions, remember context, and continuously learn from experience.
#AGI #AI #Agents #WorldModels #AIMemory #DeepLearning

[Mason Ba @PhaseNO]

Do less, so you can do it better.
Founders: In the early days, narrow your focus. Tiny scope, 100% of your energy. The world doesn’t need more mediocre stuff. It craves a few truly great things. That takes ruthless focus. Do less, so you can do it better.
#DoLess

[Mason Ba @PhaseNO]

CEO/COO/CMO must personally close the first three customers in every new region. Talk to users yourself. Only after you truly get the market should you hire local talent.Otherwise you’re not building a team, you’re throwing people into fog.

[Mason Ba @PhaseNO]

Local hires can’t replace founders for the first 3 customers.Biggest mistake when going international: thinking a local employee = “market solved.”Locals are great at scaling a market you already understand — but terrible at defining it for you.

[Mason Ba @PhaseNO]

Is the cost of the worst-case wrong answer high? If yes, you want an automation, not an agent.

Will compliance ever look at this? If yes, automation. Full stop.

[Mason Ba @PhaseNO]

If you're a founder about to spend money on an agent, answer these on paper first:
Can I draw the workflow as clear steps? If yes, you want an automation.
Does the workflow have more than five branches with truly unpredictable inputs? Then maybe an agent.

[Mason Ba @PhaseNO]

AI agents are becoming the new applications.

But most teams are still building them like scripts connected to tools.

That works for demos.

It breaks in production.

If agents are the new applications, companies need a new intelligence layer underneath them.

[Mason Ba @PhaseNO]

Hiring doesn’t replace market understanding.

If the questions are still unclear, hiring only turns uncertainty into fixed cost.
#Startups #GoToMarket #StartupGrowth

[Mason Ba @PhaseNO]

When you have to start a local team, don’t rush.
Before a market reaches $1M ARR, keep the motion lightweight:
Run content, outbound, community tests, partner channels, user interviews, remote demos, and short business trips. ↓↓↓

[Mason Ba @PhaseNO]

But don’t open an office, hire a full team before the market is validated.
Ask four questions first:
Are real users showing up repeatedly?
Do you know why they buy or don’t buy?
Can local leads be converted reliably?
Has the founder personally spoken with the first customers?
↓

[Mason Ba @PhaseNO]

@sama understanding the human beings-the body and the mind.

[Mason Ba @PhaseNO]

Exactly

[Jeff Clarke @JClarkeatDell]

The companies that win the next decade will be AI-native.
They won’t just use AI. They'll be built on it.
https://t.co/mvsYRg52X2