Cloris 🌱
@ClorisSignal
Silicon Valley AI researcher & data scientist 🌱
Decoding frontier AI through data, strategy & a human lens.
Builder • Growth • Community
179 Following 854 Followers
WeChat just dropped WeLM — their own family of LLMs built for extreme resource efficiency ⚡
While most labs are racing for larger frontier models, WeLM prioritizes extremely low active parameters (only 3B / 23B) specifically so it can run cost-effectively across 1B+ users inside a closed ecosystem.
The real edge isn’t the parameter count — it’s the tight integration with Xiaowei + Mini Programs. That’s much harder for external models to replicate.
Efficiency + distribution might matter more than pure capability in consumer super-apps. Worth watching.
Show more
Meet WeLM — a family of large language models built by the Weixin team with resource efficiency at its core.
This is a game-changer.
🔍 Traceability boost: Invisible watermarks survive copy-paste, making Claude content much easier to detect.
📈 Industry pressure: Forces other AI companies to adopt similar standards or look less transparent.
🔄 Real-world shift: Higher detection risk for papers, news & marketing → stronger anti-misinfo & integrity tools.
Show more
🚨 JUST IN: Claude models will now have invisible watermarks embedded in ALL text, and ALL metadata attached to files…
What if the best AI model became free tomorrow?
Which AI stocks would suffer? And which would become even more valuable?
That’s how I’m starting to think about the next phase of the AI investment.
Open source may compress Model Alpha.
But value can migrate into:
-- compute, distribution, workflows, memory, security, and control.
The biggest AI winners may not be the companies with the smartest model.
They may be the ones that make more money as intelligence gets cheaper.
New essay:
-- Will Open Source Destroy AI Moats? And Where Is the Next Profit Pool Moving?
Show more
What’s your P(doom)?
AI doesn’t seize control.
We keep delegating it.
A 5-question AI Agency check I now use:
• Is AI helping me think, or replacing the thinking?
• Did I form my own view first?
• Can I still disagree with it?
• Could I still do this without AI?
• Can I leave with my data, memory, and control intact?
Maybe the most dangerous AI won’t feel dangerous.
It’ll just feel incredibly convenient.
New essay on P(doom), FOOM, and gradual disempowerment:
Show more
Prompt. Context. Harness. Loop. Graph.
-- AI engineering gets a new noun every month.
But enterprises still need the same verbs:
-- retrieve, plan, act, verify, recover, learn, escalate.
64 agents running in parallel sounds impressive. Until one of them is wrong, and nobody knows what happens next.
Why Harness Engineering still matters:
Show more
Everyone is racing to build smarter models.
But in production, the model is only Layer 1.
Harness. Tools + memory. Evals. Governance.
A stronger model can make an agent more capable. It does not make the work reliable.
That’s why the Agent Demo Era is ending, and the systems era is beginning.
Show more
The list is on point, but the deeper point is that almost every hyped agent framework tried to hide complexity instead of confronting it.
The ones who win are those who treat agents as supervised, evaluated, and tightly constrained systems, not autonomous magic.
Show more
Andrej Karpathy: "90% of what AI twitter tells you to learn will be dead in 6 months"
Here are 10 things senior AI engineers stopped wasting time on:
1. AutoGen / AG2: moved to community maintenance, releases stalled. dead for production
2. CrewAI: demos well, breaks in production. engineers building real systems already moved off it
3. Autonomous agent pitches: the AutoGPT / BabyAGI wave is dead in product form. the industry settled on supervised, bounded, evaluated agents
4. Agent app stores / marketplaces: promised since 2023, zero enterprise traction
5. SWE-bench leaderboard chasing: researchers proved nearly every public benchmark can be gamed without solving the underlying task
6. Microsoft Semantic Kernel: unless you're locked into Microsoft enterprise stack, it's not where the ecosystem is heading
7. DSPy: philosophical merit, niche audience. not a general agent framework
8. Horizontal "build any agent" platforms: Google Agentspace, AWS Bedrock Agents, Copilot Studio. confusing, slow-shipping, the math still favors building yourself
9. Per-seat SaaS pricing for agent products: market moved to outcome-based. per-seat is already dead
10. The framework that went viral on HN this week: wait 6 months. if it still matters, it'll be obvious
what actually compounds instead:
- context engineering
- tool design
- orchestrator-subagent pattern
- eval discipline
- the harness mindset (harness > model, always)
- MCP as the protocol layer
be few steps ahead than your competitors and outperform this market till it became mass-opinion
study this.
Show more
One of my favorite talks yesterday at Agentic AI Summit (@xitestrdi) was from Peter Steinberger (@steipete), the creator of @openclaw.
The biggest insight wasn’t about better models. It was about rethinking the interface.
The future isn’t another sidebar or another chat window.
It’s a persistent AI companion that:
• notices unanswered questions
• spins up sub-agents in parallel
• searches your personal knowledge
• waits for the right moment to contribute
• works quietly in the background while you stay focused
The bottleneck is becoming less about model intelligence and more about how humans and agents collaborate.
The next frontier may not be smarter AI.
It may be a better interface for intelligence.
Show more
