andrea bonarrigo @WaverAndrew

22 🇮🇹| FDE-ing Chips, Post-Training @PalantirTech, @MIT | prev. Amazon andreabonarrigo.com 🇮🇹/🇬🇧/🇺🇸 Joined December 2020

Tweets

318
Followers

336
Following

260
Likes

618

andrea bonarrigo @WaverAndrew

3 days ago

welcome to world of fde-ing :)) for the same reason no one really has a good answer to broad questions like “what is the best thing you leart this year” Unless the question is specific and applied, the answer is non-significant. And simply observing past expert traces is little help either if you can’t contextualise their first principles. The beauty of asking the right questions, and setting the right vision

0 0 3 183 0

View Details

andrea bonarrigo @WaverAndrew

3 days ago

@chadwahl with all that happened after that, it has definitely been the longest-feeling, most dense 3 months ever. Onto the next 3 :)

2 0 17 606 1

View Details

Palantir @PalantirTech

3 days ago

THE SOVEREIGN AI REVOLUTION IS IN FULL FORCE. We are helping our customers build superintelligence they own, with data they control. McCarthy Building fine-tuned Nemotron in Foundry on labeled RFI data in the Ontology to identify project risks before they escalate, outperforming frontier models at significantly lower cost. Tampa General is using sovereign AI to better protect patient data, keeping sensitive data and model processing local, and fine-tuning Nemotron models to classify LLM traffic against its own risk and privacy standards. NVIDIA is using Palantir AIP and Nemotron to accelerate its own supply chain, shortening the path from wafer to deployment, spotting disruptions before they hit production, and keeping its models under its control. L3Harris is bringing AI workflows in-house and building its own models trained on proprietary data to support critical defense use cases and the warfighter. Ondas is launching Regna AI on Palantir AIP, leveraging open-weight models including NVIDIA Nemotron, to help customers adapt, evaluate, and deploy mission-specialized AI while retaining control of their data, models, and infrastructure.

52 137 998 69K 162

View Details

andrea bonarrigo @WaverAndrew

4 days ago

interview time at GTC berlin! New pareto efficiency frontier for speed and accuracy in REAL enterprise workflow is dropping!

1 2 13 1K 1

View Details

andrea bonarrigo @WaverAndrew

5 days ago

Same trend here! There is decreasing marginal return to raw intelligence now, and demand for more mindful and controlled deployments - sovereign - is all time high!

Russell Kaplan @russelljkaplan

5 days ago

This is directionally consistent with our internal data @cognition. Many tasks within software engineering are now "intelligence saturated." Once any model can do the task well, cost and speed are all that matter.

77 75 1K 154K 406

1 0 7 394 1

View Details

andrea bonarrigo @WaverAndrew

6 days ago

We’re living in a paradoxical AI populism, projecting onto AI the solution to everyone’s despair. And it’s always hard to argue against a narrative that leans into people’s needs and perpetual hope for a brighter, grander future. But reality is ultimately the most humbling force, and it never spared no one.

Lisan al Gaib @scaling01

a week ago

Surveyed Anthropic staff's answers imply 17.1% GDP growth in 2030

55 43 787 529K 249

0 1 2 214 1

View Details

andrea bonarrigo @WaverAndrew

7 days ago

Another byproduct: a creative person will always find a way to escape the boundaries it gets forced upon, give them a solution shaped box and you’ll have yourself a generational improvement

0 0 0 23 0

View Details

andrea bonarrigo @WaverAndrew

7 days ago

Human creativity, when narrowly directed, is one of the most powerful forces ever. Genius is the ability to understand what creative efforts should be directed toward.

2 0 1 137 0

View Details

andrea bonarrigo @WaverAndrew

7 days ago

It still surprises me how a technology’s popularity and availability acts as a forcing function that directs creative efforts, enabling humans to invent use cases and solutions that have lain hidden in plain sight for decades. LLMs broadly, and decision models specifically, show what happens when people collectively focus on implementing a specific tech: the fewer the strength and target focal points, the clearer and more applicable the outcomes, and problems that have sought solutions for years suddenly appear solvable.

0 0 0 103 0

View Details

andrea bonarrigo @WaverAndrew

a week ago

The boundary of capability in history has always been trust. Defining trust as confidence in a system to behave as expected under predefined conditions, trust is the single enabler that gets people on planes and makes drugs effective: the foundamental comprehension of the first principle of a technology. AI didn’t escape that pattern, and people fear inversely proportional to their understanding, which is remarkably lower then similar breakthroughs had in the past. Understanding neural networks is the single biggest enabler that will commoditize intelligence and give birth to fully autonomous workshops. Let the human understand, so that he can be freed.

Nick @nickcammarata

a week ago

interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came

878 3K 20K 4.5M 20K

0 0 1 327 0

View Details

Andrew M. Bailey @resistancemoney

2 weeks ago

Who is AI coming for next? My hypothesis: the rest of the g-loaded/verifiable quadrant. How the others fare is an open question.

126 59 705 240K 894

View Details

andrea bonarrigo @WaverAndrew

2 weeks ago

In a world in which rationality lives in a box, scaling rationality in not what bounds value creation, like it used to be; but rather figuring out where the irrationality is and leaning into its chaos. And the highest source of entropy is, and will be, the human more and more. The essence of the FDE, the living bridge between chaos and reason.

0 0 1 124 0

View Details

andrea bonarrigo @WaverAndrew

2 weeks ago

$PLTR is the only place where you go from research preview to real life testing in the same day. The ultimate RL environment for what creates values and what doesn’t.