Evan Crothers

AWS Canada Public Sector | Lead AI/ML Consultant | PhD AI/ML

[Evan Crothers @ecrws]

Bruce Schneier just called our survey on threats of machine generated text "a solid grounding amongst all of the hype."

Read it here:
https://t.co/lfqnQ7tNHk

[Schneier Blog @schneierblog]

Threats of Machine-Generated Text https://t.co/4A06RJ6msQ

[Evan Crothers @ecrws]

@TheCinesthetic Subspecies II wins this for me.

The beheaded villain simply pops his head back on, stands up, and resumes chasing the heroes -- now played by completely different actors.

Truly a masterpiece.

[Evan Crothers @ecrws]

@RealAlbanianPat Large numbers of uninformed people overreacting to new info is basically a selling feature for a prediction market.

Polymarket isn't trustable, but for other reasons.

[near @nearcyan]

surely theres a pt tho where even the unabashed techbro is like yeah the ghibli artists making US minimum wage deserve something for this

[Evan Crothers @ecrws]

@shauseth reminds me of the early days of Tesla autopilot when people would ignore every warning, find a way to let go of the wheel, and trust it with their lives.

granted the branding "autopilot" created an expectation there

[Evan Crothers @ecrws]

@shauseth vibe coding has very fixable problems to be honest. problem is that people are willing to trust AI systems 100% when they're nowhere near mature enough to justify it.

[Evan Crothers @ecrws]

This is a major opportunity for improving systems that are buckling under their weight using tools we've had for a long time now.

[Build Canada @buildcanada]

🇨🇦 healthcare system is stuck in the last century.

In ON, doctors send 150M faxes (faxes!) a year. 1 in 10 go to the wrong person. People die everyday because their records are unavailable.

Your health records should be available instantly to you and any doctor that needs it.

[LaudablePractice @cath_cov]

'Sob story'

[Evan Crothers @ecrws]

@nrehiew_ how dare you

[Nicholas Fabiano, MD @NTFabiano]

The irony

[Evan Crothers @ecrws]

@molly0xFFF Increasing user counts (and thus market cap) makes this much worse. Once there's enough money at stake, the economic incentives become incredibly dangerous.

[Molly White @molly0xFFF]

what could go wrong with running betting markets like this?

[Aidan Gomez @aidangomez]

Today, we’re excited to release Command R7B, the smallest, fastest, and final model in the R series.

It’s a powerhouse with a minimal footprint.

[Evan Crothers @ecrws]

@nearcyan Oh I know! I'll open a nuclear power plant in my backyard.

[Evan Crothers @ecrws]

@amasad I need a pedal under my desk to give the LLM more gas.

[AI at Meta @AIatMeta]

To support innovation in computer vision, we’re making DINOv2 available under the Apache 2.0 license + releasing a collection of DINOv2-based dense prediction models for semantic image segmentation and monocular depth estimation.

Try our updated demo ➡️ https://t.co/ctwiBFWS4W

[Evan Crothers @ecrws]

@Jonathan_Blow Signalis

[Riley Goodside @goodside]

Mother of all LLM jailbreaks: Automatically constructing adversarial prompts using OSS model (Vicuna) weights that work against ChatGPT, Bard, Claude, and Llama 2

Screenshots: Demo of response without/with jailbreak suffix

Linked thread from lead author has details/PDF

[Peyman Milanfar @docmilanfar]

“While the problem we address is not new, our approach to solving it is completely novel.”

[Evan Crothers @ecrws]

Good call.

https://t.co/5IuOnFjda5