iiinova — where your digital self keeps evolving

Reposted by Emmy Probasco

Ben Murphy @murphyben.bsky.social · 17/02/2026

One point I want to stress from my @csetgeorgetown.bsky.social colleagues' new report today on PLA AI procurement (cset.georgetown.edu/publication/...): Internal security is #1. Read about the AI systems China is buying to surveil its own population and to track its soldiers' activity online.

cset.georgetown.edu

China’s Military AI Wish List | Center for Security and Emerging Technology

This report examines thousands of Chinese-language open-source requests for proposal (RFPs) published by the People’s Liberation Army between January 1, 2023, and December 31, 2024. The RFPs the autho...

033

Reposted by Emmy Probasco

Lane Greene @lanegreene.bsky.social · 08/08/2025

OK, counterpoint alert: I've just spent a profitable 20 minutes chatting with GPT-4.5 in Catalan, where it has corrected my errors, explained things and gave me a quiz. It's certainly not "PhD-level" but it was *great*. It's the kind of use case these things are really for.

331

Reposted by Emmy Probasco

Dustin Moskovitz @moskov.goodventures.org · 05/06/2025

Dario Amodei has an important op-ed in the NYT today about the proposed 10-year moratorium on states regulating AI. The moratorium sounds reasonable at first - avoiding a patchwork of state laws - but it's actually quite dangerous given how fast AI is moving. www.nytimes.com/2025/06/05/o... 1/6

nytimes.com

Opinion | Anthropic C.E.O.: Don’t Let A.I. Companies off the Hook

34720

Emmy Probasco @emmyprobasco.bsky.social · 23/05/2025

Just got this from my mom. My work here is done.

010

Emmy Probasco @emmyprobasco.bsky.social · 19/05/2025

Something to consider when trying to use an LLM to summarize intel reports?

000

Emmy Probasco @emmyprobasco.bsky.social · 19/05/2025

For those looking to use LLMs to summarize scientific (in this case medical) papers: "LLM summaries were twice as likely to contain generalized conclusions compared to the original abstracts, indicating an algorithmic overgeneralization tendency." pmc.ncbi.nlm.nih.gov/articles/PMC...

pmc.ncbi.nlm.nih.gov

Generalization bias in large language model summarization of scientific research

Artificial intelligence chatbots driven by large language models (LLMs) have the potential to increase public science literacy and support scientific research, as they can quickly summarize complex sc...

110

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

There are lots of shades of gray between a coach and enforcer, I acknowledge, but regardless we must ask: Should AI override humans in combat? Could it prevent war crimes like My Lai — or create new dangers?

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

Outside the U.S., enforcer-style AI may come sooner. China and Russia are exploring AI to make up for perceived weaknesses in junior leadership.

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

Alternatively, a leader could deploy an AI as an enforcer in a way that might prevent or stop the next My Lai. But by using AI as an enforcer, they are diminishing the agency of the operators in the field.

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

A leader could deploy AI as a coach to try to prevent a violation of the laws of war, but that coach could be ignored and the leader may be criticized for not taking more decisive action.

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

Choosing how to deploy AI: as a tool, coach, or enforcer is fundamentally a leadership issue. And leaders should think about the ethical dimensions of their choices. For example:

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

But AI could also be deployed as an 🚓 enforcer: preventing or overriding human decisions when they run contrary to some specified goal.

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

AI can also be thought of as a 🏋️ coach. Especially when it’s able to fuse information or interact with natural language to help an operator achieve a goal (for an easy comparison, think of something like an advanced nutrition or diet app on your phone).

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

The thought experiment makes us consider how AI can be deployed. Most of the time, AI is thought of as a 🛠tool. A relatively limited assistant that can, for example, recognize objects and bring them to the attention of an operator.

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

Imagine a "Thompson drone" — flying overhead, spotting civilians, and warning troops in real time using voice alerts or text messages. This idea is an emerging technological possibility.

100

Emmy Probasco @emmyprobasco.bsky.social · 16/05/2025

In 1968, U.S. soldiers massacred hundreds of civilians in My Lai, Vietnam. The killing was interrupted by helicopter pilot Hugh Thompson Jr. If pilots are to be replaced by AI, then we should ask, could AI on a drone do the same in a future war?

100

Emmy Probasco @emmyprobasco.bsky.social · 13/05/2025

Wrote a thing on value-based choices when deploying AI. It's based on a military situation (a sci-fi version of the My Lai massacre) but I can't help but think this is a problem lots of leaders will face in more subtle, and less stressful, ways.

011

Reposted by Emmy Probasco

CSET @csetgeorgetown.bsky.social · 09/05/2025

As military AI evolves, ethical dilemmas grow. Should drones merely inform soldiers, coach decisions, or enforce rules of war directly? Read the new piece by Minji Jang & CSET's @emmyprobasco.bsky.social, out today in @warontherocks.bsky.social 👇

011

Reposted by Emmy Probasco

Niall Firth @niallfirth.bsky.social · 07/05/2025

Bradford G. Smith’s case highlights key issues about the prospect that brain implants and AI will one day merge. www.technologyreview.com/2025/05/07/1...

technologyreview.com

This patient’s Neuralink brain implant gets a boost from generative AI

A patient who types with his brain thanks to a Neuralink implant is using AI chatbots to speed things up.

013

Reposted by Emmy Probasco

Tim G. J. Rudner @timrudner.bsky.social · 01/04/2025

I'm super excited to see our #CSET report on **AI-enabled military decision support systems** being released today! Great work by @emmyprobasco.bsky.social, @hlntnr.bsky.social, and Matthew Burtell!

001

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

With all this new data from remote sensing, new space data analysis companies are also starting.

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

And new companies are also seizing the opportunities presented by new sensing modalities:

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

It also helps that camera tech is improving for these remote sensing providers (thanks to government R&D!)

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

It’s not just launch that’s different: take a look at how many remote sensing companies have popped up since 2010. Helps that these guys have cheaper launch options:

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

These companies are making faster and faster progress towards stable launch capabilities:

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

The good news is that there has been a raft of new launch companies founded since the 2000s.

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

Today, the players have changed dramatically. SpaceX can be credited for revolutionizing launch and the commercial space industry, but SpaceX’s dominance raises questions about resiliency too.

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

Also, the space launch industry in Reagan’s time was dominated by the defense industrial base:

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

Zooming out to include modern times, we’ve come a long way…

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

Launch (thanks largely to commercial launches) is dramatically different. Reagan announced SDI in 1983, three years later the Challenger exploded before the eyes of Americans and put a chill on launches for years following.

100

Emmy Probasco @emmyprobasco.bsky.social · 31/03/2025

There are plenty of technical & financial hurdles to defense against “ballistic, hypersonic, and cruise missiles, and other advanced aerial attacks” #GoldenDome. But one thing is very different from Reagan’s days: the commercial space industry. A story in @csetgeorgetown.bsky.social charts: 🧵

100

Reposted by Emmy Probasco

CSET @csetgeorgetown.bsky.social · 20/03/2025

In this month’s newsletter, CSET's @alexfriedland.bsky.social wrapped up all the latest AI & policy news: 1️⃣ GPT-4.5 & Claude 3.7 Sonnet 2️⃣ Recapping new misalignment research 3️⃣ DOD wants to speed up software acquisition 4️⃣ AI Action Plan recs And much more! cset.georgetown.edu/newsletter/m...

cset.georgetown.edu

OpenAI and Anthropic try to fend off competition with new models, ideas for the U.S. AI Action Plan, and important new misalignment research | Center for Security and Emerging Technology

Plus: DOD looks to accelerate software acquisition, Trump floats CHIPS repeal, White House tries to cut off maintenance of Chinese chip tools, and the confirmation hearing for new OSTP head

042

Reposted by Emmy Probasco

Dewey Murdick @deweyam.bsky.social · 17/03/2025

The future of AI leadership requires thoughtful policy. @CSETGeorgetown just submitted our response to @NSF's RFI on the Development of an Artificial Intelligence Action Plan. Here's what we recommend: 🧵[1/]

155

Emmy Probasco @emmyprobasco.bsky.social · 10/03/2025

Had fun chatting with Gen. Tim Haugh (US CYBERCOM/NSA about military applications of AI in Austin this Friday. Video ICYMI. www.youtube.com/live/yIMRxOg...

youtube.com

Joint Disruptive Defense (March 7)

YouTube video by Austin Tech Live

021

Emmy Probasco @emmyprobasco.bsky.social · 03/03/2025

AI could both harm and help in balancing efficiency with thorough analysis. While AI often drives efficiency, it could also be configured to force consideration of alternative explanations, enhancing the effectiveness of intelligence analysis. This balance is fundamentally a human choice.

000

Emmy Probasco @emmyprobasco.bsky.social · 03/03/2025

Data management issues in AI systems could mirror those in traditional intelligence, including challenges like ensuring the relevancy and accuracy of data, adequate sensor deployment, and appropriate metadata and ontologies for decision-making.

100

Emmy Probasco @emmyprobasco.bsky.social · 03/03/2025

There is evidence that AI can change a person’s mind, but if there's a prevailing belief that certain scenarios are unlikely, AI might not be able to overcome that bias. arxiv.org/pdf/2403.14380

100

Emmy Probasco @emmyprobasco.bsky.social · 03/03/2025

While AI doesn't directly address the decline in traditional surveillance methods like HUMINT, it could potentially lower the costs and increase the efficiency of OSINT and automated analysis of communications.

100

Emmy Probasco @emmyprobasco.bsky.social · 03/03/2025

AI could potentially enhance the critical nature of the intelligence work process, especially if implemented as a red team to identify and challenge assumptions (as seen in IARPA’s project REASON) www.iarpa.gov/research-pro...

100

Emmy Probasco @emmyprobasco.bsky.social · 03/03/2025

AI could exacerbate the perception that the enemy is transparent when it really is not. Especially problematic when relying on large datasets that require context and understanding.

100