Raj (@[email protected])

Director of ML at @deepsignl

[Raj (@[email protected]) @rajb245]

@jormungandrizer @RetroMorgan89 @retrotink2 I’m going mostly with mister for the things it supports, and then working my way through acquiring the newer stuff (6th gen and up, already have the Xbox and GameCube). And rt4k for everything into an LG OLED.

[DeepSig @deepsignl]

OCUDU redefines how #wireless networks are built, deployed, and evolved. By combining carrier-grade CU/DU software, multi-vendor support, accelerated compute, and AI-native design, #OCUDU unlocks faster innovation, greater customization, and deeper ecosystem collaboration across public and private networks.

Read more about how OCUDU is shaping the next era of wireless innovation: https://t.co/IjtcpAxEAP

#AI #5G #6G #WirelessInnovation #OpenSource

[Raj (@[email protected]) @rajb245]

@RetroMorgan89 @retrotink2 But makes me wonder what’re you using for the RGB consoles, a retrotink?

[Raj (@[email protected]) @rajb245]

@RetroMorgan89 @retrotink2 Kind of funny to show a bunch of HDMI mods to Mike, who literally makes the best thing you can buy to NOT have to HDMI mod any of your consoles lol

[Raj (@[email protected]) @rajb245]

The hardware changes down the stairs while the red words are applied to one model to find weights. Not the tightest analogy, but suggests GA/EA/gradient free searches for architectures could learn analogously.

[François Fleuret @francoisfleuret]

"the human brain doesn't need tons of training data to do stuff"

[Raj (@[email protected]) @rajb245]

@NotBaeM2

[DeepSig @deepsignl]

We’re headed to #GRCon25 this week, the place where the open source and #SDR community comes together to share, learn and innovate.

#DeepSig will be showcasing how our AI-native software OmniSIG® accelerates #spectrum awareness. From faster and smarter signal detection with Python and C++ support, to turning unknown #signals into known ones in hours with OmniSIG Studio, to #AI-learned waveforms that adapt to real-world environments, #OmniSIG delivers practical performance where it matters: https://t.co/yjZlXIP1Nz

[Raj (@[email protected]) @rajb245]

It’s the speakership, not the vice presidency. “But no person constitutionally ineligible to the office of President shall be eligible to that of Vice-President of the United States.” But anyone can be third in line, speaker of the House.

[Raj (@[email protected]) @rajb245]

@ezyang Do you know if it’s the helpful assistant fine tuning that gives the model a preference for markdown? Seems the base model before that shouldn’t have a bias towards markdown because lots of the internet isn’t markdown

[Light Reading @Light_Reading]

Startups are tinkering with Aerial, Nvidia's RAN software, and the results could shake up the industry.

Read more on Light Reading: https://t.co/yv5GZCQ6YN

[Raj (@[email protected]) @rajb245]

@ATunnelrat The VA got hit too

https://t.co/K2Y5PLwjwO

[ATourist @ATunnelrat]

I am a huge supporter of trump and this administration. I also support Doge and am a huge fan of Elons. What I don’t support is the attempt to fire all probationary fed employees, and the freeze of all placement, promotions, and laterals of federal employees. Cont.

[Raj (@[email protected]) @rajb245]

@realGeorgeHotz @blitz_or_die @comma_ai @LisaSu TinyCorp hasn’t made AMD not suck for ML workloads yet?

[Raj (@[email protected]) @rajb245]

@minimaxir How. Dare. You. Sir.

[Raj (@[email protected]) @rajb245]

This is also true in RF, we haven’t cracked unsupervised pretraining. Also no one creates dozens of RF recordings and uploads them to social media daily BUT the chipsets could certainly capture this data for us

[Dhruv Batra @DhruvBatra_]

Brilliant talk by @ilyasut, but he's wrong on one point.

We are NOT running out of data. We are running out human-written text.

We have more videos than we know what to do with. We just haven't solved pre-training in vision.

Just go out and sense the world. Data is easy.

[Raj (@[email protected]) @rajb245]

@HanchungLee @jeremyphoward I second this one. A docker tag in a registry per CUDA/python version combo you want to support, and -v mounting the source you’re working on to inside the docker. Spin up container, build and test. Spin up another container, build and test.

[Raj (@[email protected]) @rajb245]

@karpathy They do seem to be very good at manipulating tokens in symbolic math, and come up with answers far beyond what the labeler could have done. For example, write this integral, substitute some other representation, simplify; it applies identities the average labeler wouldn’t know

[Raj (@[email protected]) @rajb245]

@robertghilduta What’re you up to…is it RF/signals related or something else?