🔭 How NASA Used AI to Find Possible Signs of Life on Mars

While Congress debates whether or not there’s alien life on earth, NASA may have just found our first tangible evidence of ancient life on Mars. NASA’s Perseverance rover cored a sample called “Sapphire Canyon” from a rock nicknamed Cheyava Falls in Jezero Crater’s Bright Angel formation. 

In a peer-reviewed Nature paper and NASA’s release, the team describes mineral and organic patterns: “leopard spots,” iron phosphates (like vivianite), and iron sulfides (like greigite). On Earth, these often coincide with microbial activity. NASA calls them potential biosignatures — promising, but not a life detection. The claim needs more tests and, ideally, Earth-lab analysis. 

🧠 Where AI Actually Helped

AI didn’t “declare life.” It enabled the hunt. Perseverance’s AutoNav self-driving maps terrain and plans paths, which lets the rover reach difficult outcrops faster and safer than fully tele-operated drives. Its AI targeting software AEGIS can autonomously pick and zap promising spots for instruments between ground contacts, boosting science in tight comms windows. 

The PIXL instrument’s adaptive sampling uses machine-learning rules to spend more time on chemically interesting pixels, raising signal-to-noise exactly where it matters. Together, these systems made it more likely the rover would arrive at, notice, and characterize an unusual rock in a short Martian field season.

🧪 What the Evidence Is (And Isn’t)

The textures and chemistry could come from ancient metabolisms, or from non-biological geochemistry that mimics them. That’s why the phrasing matters: potential biosignature. 

Perseverance’s in-situ tools (for example, the PIXL X-ray microprobe and the SHERLOC Raman/UV spectrometer) point to intriguing combinations, but the decisive tests (isotopes, microstructures in thin section, contamination controls) are best done in Earth labs. NASA’s Mars Sample Return path remains the gold standard, even as timelines and budgets evolve.

🧭 Why This Is Still a Big Deal

AI-driven autonomy is shrinking the gap between what scientists wish they could sample and what a rover can actually reach, inspect, and store. 

💢 The upshot: more targeted cores in the cache, and higher odds that one carries a true biosignature, whenever it gets home.

✔️ Bottom line: AI didn’t “find life.” It raised the odds of finding the right rock, then helped instruments squeeze more truth from it, an assist that could make all the difference when proof requires patience.

🏛️ Inside California’s Landmark AI Safety Bill

California’s SB 53 targets frontier models, the most powerful general-purpose AI systems, with a transparency-first approach. Rather than dictating architectures, it would require the biggest developers to publish safety frameworks, report critical incidents, and preserve whistleblower protections. The goal: “trust, but verify” for labs whose systems could, in worst-case scenarios, drive serious public-safety harms. 

📜 What’s in the Bill

Expect mandated public safety reports describing catastrophic-risk mitigations, incident disclosure to state officials, and guardrails around data retention and internal escalation. Staff who flag risks gain protections against retaliation. Analysts note the bill intentionally stops short of the most controversial SB 1047 ideas (like mandated kill-switch audits), trading breadth for a clearer compliance path that large labs can execute.

🤝 Who’s Backing It & And Why

Anthropic publicly endorsed SB 53, framing it as a pragmatic, state-level baseline while Washington dithers. Civic groups backing transparency and youth safety are on board, while parts of the industry remain wary of state patchworks and compliance costs.

Politically, SB 53 gives Sacramento a way to claim leadership without repeating last year’s vetoed approach, while testing whether voluntary pledges can be codified.

🔮 What to Expect Next

If SB 53 passes a final vote and reaches the governor’s desk, it could become the first U.S. state law specifically aimed at frontier-model transparency and risk reporting. Even before enforcement, large buyers (cities, universities, agencies) may cite its disclosures in procurement, nudging a de facto standard.

Watch for clarifying regulations on who qualifies as a covered developer, what counts as a “critical incident,” and how California will publish (and police) these reports.

Bottom line: SB 53 won’t solve every AI harm, but it may lock in the minimum receipts: evidence, not promises, from the labs building the sharpest tools.

This Week’s Scoop 🍦

🧠 Ex-Google X trio raises $6M to build a “second brain”

📚 Eliezer Yudkowsky’s AI book gets the Times treatment

🤝 Microsoft may diversify beyond OpenAI with Anthropic buys

🏙️ Trump posts AI image touting Chicago crackdown

🏛️ BNY Mellon to increase funding in Pennsylvania's energy and tech

🛡️ Publishers test a protocol to block unwanted AI crawlers

🎯 Weekly Challenge: Ask AI Better Questions

Prompting isn’t dead, it’s evolving, and so should we. This week, we’ll learn how to ask AI specific questions to get more aligned answers.

✍️ TLDR; Paste ready template
“Goal [one sentence]. Role [who you are]. Audience [who this is for]. Constraints [tone, length, format, avoid]. Inputs [background]. Produce two options. Then give a three line approach, one key assumption, and a confidence rating. If a fact is uncertain say unknown rather than guessing.”

Follow these steps on any topic you care about and save your best results as a reusable template:

📌 Define the outcome
Write one sentence that states the goal and a success metric, like “Create a 150 word intro that earns two replies.”

🧑‍🏫 Set the role and audience
Tell the model who it is and who it is writing for, like “You are a career coach writing for new grads.”

🧱 Add constraints
Specify tone, length, format, and no go items, like “friendly, 120 to 160 words, three bullets, avoid clichés.”

🔎 Require evidence or limits
Ask for named sources when facts matter and instruct the model to say “unknown” if it is not sure.

🧰 Give inputs
Paste context, examples, or data. Label sections clearly, like “Background” and “Must include.”

🧪 Request two distinct options
Ask for Option A and Option B that differ in structure or angle so you can pick a winner.

🧭 Add a quick self check
Ask for a three line summary of approach, the top assumption, and a confidence rating.

✏️ Iterate once
Tell the model to fix two issues you notice and to shorten or clarify by ten percent.

🧾 Save your template
Keep a version with blanks to fill next time.

Want to sponsor Jumble?

Click below ⬇️

If AI can help us find life on Mars, can it also help us figure out what’s flying in our skies here on earth? And, is SB 53 a step in the right direction or will it hurt the US in its race for AI supremacy? We’d love to hear your thoughts. See you next time! 🚀

Stay informed, stay curious, and stay ahead with Jumble!

Zoe from Jumble