Bertrand Schmitt

EIR at Co-Host Co-Founder @AppAnnie/@DataAI, @ISEP and @Wharton alum, 🇫🇷 born and raised. Tweets are my own

[Bertrand Schmitt @bschmitt]

@ptsi Pleasure, was reading some of your posts on browser control with AI, thanks for the content!

[Bertrand Schmitt @bschmitt]

@noguchis Excellent leveraging some ideas from your version, I am getting 50 t/s on RTX Pro 6000

[Bertrand Schmitt @bschmitt]

@antirez Yes 96GB of VRAM

[Bertrand Schmitt @bschmitt]

@antirez Specs say more is possible: RTX Pro 6000 has 6.5× DGX Spark's mem BW.

nvidia-smi dmon shows the GPU's memory bandwidth peaks at only ~33% during gen, while nsys shows ~1,800 small kernel launches per generated token.

[Bertrand Schmitt @bschmitt]

Latest DS4 q2-imatrix on RTX Pro 6000 + Linux bare metal:

@ ctx 8k → 479 t/s prefill · 41.7 t/s gen
(env tweak DS4_CUDA_Q8_F16_CACHE_RESERVE_MB=512 = +33% prefill)

vs antirez/ds4 README:
DGX Spark 343/14 · M3 Ultra 468/27 · M3 Max 250/22

= 3× DGX gen, 1.4× DGX prefill.

@antirez

[Bertrand Schmitt @bschmitt]

@antirez @NVIDIAAIDev would be great to support @antirez efforts on DGX Spark and RTX Pro 6000

[Bertrand Schmitt @bschmitt]

@antirez Excellent news, thanks for the great work, can't wait to test and run it on RTX Pro 6000

[Bertrand Schmitt @bschmitt]

@antirez @rikarends 80+ tokens/s is not an unreasonable target

[Bertrand Schmitt @bschmitt]

@antirez @rikarends Actually RTX Pro 6000 memory bandwidth is 1,792 GB/s

[Bertrand Schmitt @bschmitt]

@llmdevguy @UnslothAI @Kimi_Moonshot better have a 2nd rtx 6000 pro, and very fast RAM with an EPYC CPU

[Bertrand Schmitt @bschmitt]

@Jasonch14221644 @XPlaneOfficial 100%, performance is the biggest issue at this stage for X-Plane, specially for those of us in VR

[Bertrand Schmitt @bschmitt]

@AndreasSteno Interesting reply on this exact topic https://t.co/MDlYFCQHki

[Nick Timiraos @NickTimiraos]

The first headline that is quoted in Bob's thread is not an accurate headline. Maybe it's engagement farming, maybe people just don't know better. Just because someone puts it in caps doesn't make it an accurate headline.

The Trump administration said in February it was going to seek to overturn Humphrey's, which is the legal decision upon which most scholars think Fed independence rests. Humphreys was actually decided in the spring of 1935, when the Congress was passing the Banking Act of 1935 that created the modern Fed. It established the FOMC as a monetary policy making body, which hadn't been part of the original 1913 Act. Congress relied on Humphreys when creating staggered terms for Fed governors, among other measures, that were explicitly designed to insulate the central bank from presidential control (notably, against the wishes of the then-Fed chair, Marriner Eccles, who wanted a central bank board that answered straight up to the president).

It is true that overturning Humphreys could have significant consequences for the Fed, but this is not a given and either way, Trump has asked to fire FTC commissioners and NLRB members, not Fed governors. His DOJ is challenging, head on, a decision that is widely viewed as insulating Fed governors from removal. It is possible that the Supreme Court would design a ruling that nukes the job-security protection for FTC commissioners and NLRB board members but does not do so for the Fed. Whether the Roberts Court does that is a separate question entirely.

For more on this, I highly recommend former Fed governor Dan Tarullo's recent law review article that walks through many of these issues.

To get to the point of his article, it "discusses why and how, notwithstanding these apparent constitutional vulnerabilities, the Court might well not hold the core delegation to, and structural features of, the Federal Reserve to be unconstitutional. As to why—members of the Court’s conservative majority may be more favorably inclined toward a central bank than other economic regulatory agencies. A more tangible consideration is the difficulty the Court would have in fashioning a remedy for the supposed unconstitutionality of the FOMC structure or mandate that did not risk major disruption to monetary policy, and thus the U.S. economy."

"As to how.... the Court may find that, on the merits, the Federal Reserve enjoys an exception to the doctrines the Court’s majority has been building. This second way itself has two branches. One is based on the history of the regulation of money going all the way back to the First Bank of the United States. The other rests on perceived functional differences between the Federal Reserve and other independent agencies—an 'anomaly,' as then Judge Kavanaugh once described it."

https://t.co/3kpv6U8tDC

[Bertrand Schmitt @bschmitt]

@XPlaneOfficial Excellent, I remember there was a plan for performance improvements, is it still coming? Triple screen 4K (without 3 computers) and VR still get still very poor framerates, even on the most highend hardware you can get today...

[Bertrand Schmitt @bschmitt]

@XPlaneOfficial Please performance improvements for triple screens and VR!

[Bertrand Schmitt @bschmitt]

@markankcorn @dhh Framework Desktop has to USB4 ports, compliant with Thunderbolt 4, so can go to 40gbps

[Bertrand Schmitt @bschmitt]

@neilcybart @waitbutwhy My primary reason for not using much my AVP is the lack of comfort. Even 30 min is painful. Let alone 2 hours. In comparison, my Varjo VR4 is way more comfortable, and I will use it longer.

[Bertrand Schmitt @bschmitt]

Glad to see the complexity of OpenAI offering is going to disappear

[Sam Altman @sama]

OPENAI ROADMAP UPDATE FOR GPT-4.5 and GPT-5:

We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings.

We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.

We hate the model picker as much as you do and want to return to magic unified intelligence.

We will next ship GPT-4.5, the model we called Orion internally, as our last non-chain-of-thought model.

After that, a top goal for us is to unify o-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks.

In both ChatGPT and our API, we will release GPT-5 as a system that integrates a lot of our technology, including o3. We will no longer ship o3 as a standalone model.

The free tier of ChatGPT will get unlimited chat access to GPT-5 at the standard intelligence setting (!!), subject to abuse thresholds.

Plus subscribers will be able to run GPT-5 at a higher level of intelligence, and Pro subscribers will be able to run GPT-5 at an even higher level of intelligence. These models will incorporate voice, canvas, search, deep research, and more.

[Bertrand Schmitt @bschmitt]

@yishan Great analogy!

[Bertrand Schmitt @bschmitt]

@sc_cath Also deepseek breakthrough optimizations benefit all AI Semiconductor companies, not just Nvidia

[Bertrand Schmitt @bschmitt]

Pretty good list!

[Martin Tobias (Pre-Seed VC) @MartinGTobias]

I have invested in over 250 pre-seed startups, mostly in B2B software. The most successful ones have had some distribution tricks to quickly get their first happy customers.

Here are the Best (and worst) distribution wedges I have seen for Pre-Seed software startups.