Ahmad Bader

SDE Intern @amazon | MSc Data Science @UniofGreenwich

[Ahmad Bader @baderowski]

@krishna90124608 mera number to active hai

[Ahmad Bader @baderowski]

Per 90 minutes, Messi beats Ronaldo on all five on-ball attacking metrics.

[Ahmad Bader @baderowski]

An evening of sharp talks on AI at scale with the Ring (Cambridge) team. Grateful for the invite.

[Ahmad Bader @baderowski]

Claude Sonnet 4.5 feels sentient now xD

@claudeai @AnthropicAI

[Aakash Gupta @aakashgupta]

Naval really nailed it with this one: Money < life < attention

[Paul Graham @paulg]

Hamming's talk is so important that I reproduced it on my site. It's one of the only things on my site written by someone else.

https://t.co/kWvKdwIiOm

[Ahmad Bader @baderowski]

One of the best definitions of “agency” I’ve read:

"Agency is the exercised belief that outcomes in your sphere are produced by your actions, combined with the disposition and capacity to produce those actions, especially under adverse or ambiguous conditions."

[Ahmad Bader @baderowski]

Initial LLM loss curve, might as well just use open weights xD

#LLMs #Training #gpt2

[Ahmad Bader @baderowski]

While learning with Gemini Pro 3.0, there comes a point where it just collapses and starts infinitely printing ‘pass’ for some reason :P
@Google @GeminiApp @GoogleDeepMind

[Ahmad Bader @baderowski]

@WorkaholicDavid buttons

[Ahmad Bader @baderowski]

@peach2k2 It's actually interpreted as follows:

while ((x--) > 0)

[Ahmad Bader @baderowski]

Truth above all else.

[François Chollet @fchollet]

To really understand a concept, you have to "invent" it yourself in some capacity. Understanding doesn't come from passive content consumption. It is always self-built. It is an active, high-agency, self-directed process of creating and debugging your own mental models.

[Ahmad Bader @baderowski]

Most beautiful things are inexplicable.

[Chrisman @chrisman]

There’s an essay where a famous magician says the secret to magic is doing an inconceivable amount of work for a trick. So much work that it’s easier for audience to believe you are magic.

I think this might be the key to doing anything great.

[Ahmad Bader @baderowski]

5) Takeaway:

The path forward is not brute force scaling.
It’s smarter AI:
-world models
-curated datasets
-insight over size
-etc.

#AI #machine #learning #LLM #degenerative #deeplearning #ScalingLaws #resarch #paper

[Ahmad Bader @baderowski]

3) Learning vs Stability tradeoff

Transformers thrive because they turn Gaussian inputs into non-Gaussian outputs.
But: non-Gaussian outputs = fat tails → rare, extreme errors → error pileup.

[Ahmad Bader @baderowski]

2) “The Wall” is real

Like round-off errors in numerical math, LLMs may reach a point where more compute worsens performance. Adding resources can actually increase error.

[Ahmad Bader @baderowski]

@MIT_CSAIL agriculture and construction