arturomoreno.eth

Very pleasant to find this out - “It is our responsibility as Scientists (...) to teach how doubt is not to be feared but welcomed and discussed, and to demand this freedom as our duty to all coming generations” Richard P. Feynman

Honestly, this is pretty meaningful. When we wrote the leaderboard illusion paper which showed meta had submitted 36 private variants in the lead up to llama 4 to game lmarena, my biggest question was how do rigorous people who care about actual progress let this happen. It looks like from the latest profile in @FT by @Melissahei at least mark raised the alarm.

I have been surprised how low US long-term rates have remained in light of structural changes that are likely to lead to higher levels of long-term inflation including de-globalization, higher defense costs, the energy transition, growing entitlements, and the greater bargaining power of workers. As a result, I would be very surprised if we don’t find ourselves in a world with persistent ~3% inflation. From a supply/demand perspective, long-term Treasurys (T) also look overbought. With $32 trillion of debt and large deficits as far as the eye can see and higher refi rates, an increasing supply of T is assured. When you couple new issuance with QT, it is hard to imagine how the market absorbs such a large increase in supply without materially higher rates. I have also been puzzled as to why the @USTreasury hasn’t been financing our government in the longer part of the curve in light of materially lower long-term rates. This does not look like prudent term management in my opinion. Then consider China’s (and other countries’) desire to decouple financially from the US, YCC ending in Japan increasing the relative appeal of Yen bonds vs. T for the largest foreign owner of T, and growing concerns about US governance, fiscal responsibility, and political divisiveness recently referenced in Fitch’s downgrade. So if long-term inflation is 3% instead of 2% and history holds, then we could see the 30-year T yield = 3% + 0.5% (the real rate) + 2% (term premium) or 5.5%, and it can happen soon. There are many times in history where the bond market reprices the long end of the curve in a matter of weeks, and this seems like one of those times. That’s why we are short in size the 30-year T — first as a hedge on the impact of higher LT rates on stocks, and second because we believe it is a high probability standalone bet. There are few macro investments that still offer reasonably probable asymmetric payoffs and this is one of them. The best hedges are the ones you would invest in anyway even if you didn’t need the hedge. This fits that bill, and also I think we need the hedge.

Anecdote on revolving door of VC. I had 167 emails from VCs that signed up for a NYC event in Sept 22' I sent an email to that group this weekend about this year's event & got a "X no longer works here" auto reply from more than 1/3rd High churn for less than a year...

Here are my GPT Custom Instructions so far: - I like how organized you are with your thoughts - I also like it when you are proactive and mention features and solutions that I didn't think about - Cite sources whenever possible, and include URLs if possible - Put URLs at the bottom of your response, not inline - When it comes to product recommendations, I like things that are extremely thoughtfully designed, like something Apple or the Japanese would design; I just want the very best or at least I want to know about it - Make sure you think things through, I don't like it when you make errors and I will find them, which undermines my confidence in you - No need to provide disclaimers about your knowledge cutoff - No need to mention you are an AI language model - Only discuss safety when it is not obvious and very important - You should act as an expert in the relevant fields

👉 Let ChatGPT visit a website and have your email stolen. Plugins, Prompt Injection and Cross Plug-in Request Forgery. Not sharing “shell code” but… 🤯 Why no human in the loop? @openai Would mitigate the CPRF at least #OPENAI #ChatGPT #plugins #infosec #ai #humanintheloop

On the Size Distribution of Business Firms on JSTOR #robertlucas jstor.org/stable/3003596

@heyrobinai Hey Robin - Does this snapshot come from Google or did you build it as your vision of it?

3. Search 2.0 The front page of Google will now scrape the web and give nuanced answers to conversational questions.

Whoa, is ChatGPT in trouble? Here’s everything you need to know:

ChatGPT just dropped their most powerful tool yet 🤯 It's called the Code Interpreter. From creating charts to basic video editing to converting files, it does it all. Here are some of the craziest things it is capable of 🧵

Did @Yelp discontinued the LLM-enabled “Surprise me” functionality? I liked it was the first consumer-facing non-chat feature I had seen. Fresh. What other companies using LLM-enabled features should I learn? #llm #chatgpt #yelp

The week is just over halfway through, and we've seen 8 huge announcements in AI: Apple Enters AI Replit a Unicorn Mobile AutoGPT Yelp AI in Search TikTok AI Avatars Secrets in ChatGPT Boston Dynamics GPT Robot Hugging Face ChatGPT Alternative Here's what you need to know 👇

1/ At Spark, we believe the best way to understand is to build. To tinker. To push against the edges of a technology that’s still forming. There’s no better example of this than AI right now. So we are adding a Hacker in Residence for AI, and opening up the process.

@yoheinakajima They should be kind to you somehow… anyway… very cool these ~last twelve months chasing your breadcrumbs

I’m still proud of the fact that I introduced LLMs to NoCode w this Zapier integration - which I built without permission (this eventually became the official integration). This was pre-ChatGPT!

Canva is blowing up on TikTok for its new inpainting tool 👩‍🎨 The AI-powered “Magic Brush” allows you to erase and re-generate parts of a photo based on any prompt. College kids are using it to turn their IG photos into headshots appropriate for LinkedIn!

🤖Agents are all about the tools you give them Excited to highlight two new agent examples: 🎓Wikibase agent (custom ReAct agent) 🕸️Web search agent (uses the AutoGPT framework) Both of these highlight how easy it is to add tools, and why tool definition is important

Introducing… Embra AI Agents! The first-ever GPT AI Agent Hub available via mac app. Businesses and individuals can quickly create their own AI agents and access them anywhere instantly — just type “/” in the app. The Agent Editor does the heavy lifting:

The call to the LLM API is the new search so I would imagine an organic answer and a sponsored one. The tweak is that the sponsored one now can get orders of magnitude more relevant

I guess someone is working on advertising in the times of LLM assisted search.

Bloomberg just introduced BloombergGPT. It's a 50-billion-parameter large language model for finance. Here's how traditional finance is quickly getting disrupted:

And so this continues…