Jonathan Sabbah

@kennnnchen What’s the « software factory » from the left bar?

shipped something small: • alt + click “new chat” • cmd + alt + O runs in the background. no context switching.

@ivanhzhao interested!

LMK if you want early access (and can provide ample feedback in exchange)

@MarissaJFelix @NotionAPI @attambulkar they did now, and the notion MCP is still not working (even though i could add the credentials successfully) Honestly, it’s starting to get a bit annoying, because it feels like this system isn’t very well polished yet :/ @n8n_io

@SabbahJonathan @NotionAPI @attambulkar Just took a quick look! Looks like n8n doesn’t currently support connecting to an external MCP server—only the other way around is officially supported. Would love to be wrong though. lmk if I missed something! For now, seems like this one’s on n8n’s side

Amazing use case for MCP and @NotionAPI! thank you for sharing @attambulkar!

@emollick Notion AI solves it nicely by having a prompt in a public page and in the Notion AI chat everyone can use it

Its fine if it turns out that GPTs/Gems/whatever aren't the future, but it seems reasonably urgent to roll out something else that makes sharing prompts useful across teams and organizations. Prompt libraries are still important, and they are still awkward cut-and-paste things.

I'll note again that it seems nuts that, despite every AI lab launching a half-dozen new products, nobody is doing anything with GPTs, including OpenAI When I talk to people at companies, this is still the way non-technical people share prompts on teams. No big change in 2 years

@NotionHQ I use the new webhook system to fetch posters & metadata for movies, books, series, and games based on a title. As a result, I can generate statistics on the actors I’ve watched the most among other cool things :)

🔧 Are you building something wonderful in Notion? We want to hear from you in the replies! Every Thursday, we'll spotlight the best builders on our blog. We’re talking to the doers who are making things happen with grit and passion. Does that sound like you?

@emollick @emollick I read your tweets everyday and it feels like you’re the one who can lead that project

GPQA Diamond and MMLU and ARC-AGI look nothing like actual work. I am surprised no large IT, consulting, or even national standards agency hasn't stepped in with large-scale batteries of private tests. Especially given the hundreds of billions being invested.

Another thing Grok 3 highlights is the urgent need for better batteries of tests and independent testing authorities. Public benchmarks are both "meh" and saturated, leaving a lot of AI testing to be like food reviews, based on taste. If AI is critical to to work, we need more.

@emollick Very similar findings in this MIT study from nov. "while the bottom third of scientists see little benefit, the output of top researchers nearly doubles" aidantr.github.io/files/AI_innov…

New study on AI & investing: When GPT-4o summarizes earnings calls to match investor expertise level (simpler for novices, technical for experts): Sophisticated investors get +9.6% improvements in 1-year returns, novices: +1.7% AI helps everyone, but expertise amplifies benefits!

I once read a book about Toyota’s history and how the kanban board was invented there 70 years ago. Today, Toyota is a @NotionHQ customer and uses Notion kanbans. Pretty cool!

@NotionHQ FORMS

You’ve been following along for the summer of launches, just recently Charts 📊 We have one last thing that we couldn’t keep to ourselves, so we’re rounding out the Summer of Leaks with — Forms 🚨Leaked Notion all-hand footage🚨 → Want early access? Like this and reply “FORMS 📝”

Three things about advanced voice on ChatGPT: 1) It is as good as the demo 2) It is clearly capable of producing a lot more audio (I occasionally get sound effects, etc) but there are guardrails 3) It is super weird. Lots of unconscious cues make it feel like talking to a person

@keithwhor same here

my X feed has gotten increasingly political & toxic (esp conspiracy heavy) over the last several mos. i didn’t notice until it upset me a couple days ago. if i don’t aggressively curate my mute list like weeding a garden it gets bad, fast. has anyone else noticed a change here?

From our first 1,000 users → to 100 million! We’ve loved building with all of you. 🧱 Thank you for being part of our journey and allowing us to be a page in yours.

@NotionHQ Charts 📊

Summer of launches is just getting good…Coming soon 👀 Charts! You've been entering data into Notion databases for years, enjoying the various views, filters, and connections. Now, turn that data into a story with—Charts. Rolling out later this summer… Reply “Charts 📊” to get added to the early access waitlist.

@levelsio We hear about a problem and our brain immediately jumps to a solution. Human nature. But research tells us that first ideas are poor. They are what we have in mind at the moment. Not the smartest, most original or efficient solution to a problem

The reason for this I think is: The first idea submerges from your subconcious and was actually boiling in your brain for weeks, months, sometimes years in the background The 8 hours of thinking can never compete with a perpetual subconcious background job

@keithwhor @autocode @OpenAI Congrats!!

hello! today I have some news; @autocode is shutting down and I am joining @OpenAI. announcement link in follow up; shutdown is Apr 26. the quick summary is that I got offered the opportunity of a lifetime on the API & Platform team to bring my knowledge and experience building Autocode to OpenAI. i'm incredibly excited. Autocode is my "first startup love" - our teammates, investors, customers, community have been awesome, thank you. and thank you to @threesided especially who walked with me all the way to the finish line, i'm incredibly grateful for your friendship & support. i wouldn't have this opportunity without you. if you never got a chance to play around, @autocode is available for folks to check out until Apr 26!

@eldsjal i have an "inbox playlist" to store all the albums i have to listen before putting the one i like the most in my library

I love seeing the creative ways people use Spotify. Does anyone have an unusual use case you think others should try?

@craftworkdesign Is it possible to have some insights on how my sales are going / when / how I will be paid? I know that I have customers coming from Craftwork, but I have no news from you and you have not replied to my emails, DM, message etc so i’m a little bit worried 😅

View my verified achievement from @NotionHQ. credly.com/badges/fb4e849…

@hrishioa for the deployment how about @huggingface ?

How many years until open source models take over? Is it never? I've been manually testing 20-30 different models all claiming impressive scores on benchmarks against OpenAI and Anthropic. What I've found: * Super tiny models are becoming insanely good * Medium models are starting to beat GPT-4 with prompt tuning and on specific tasks * GPT-4 still has a huge lead if you only want to use one model * Claude-instant is being slept on * OSS models have a pretty big deployment problem 1. Tiny models will definitely win: StableLM-Zephyr, at 3B (2.9 GB on disk and 50 t/s on most hardware) is punching way above its weight and providing completely working, coherent responses - all the way from writing to SQL generation. For some tasks it's indistinguishable from 3.5-turbo. At this size you solve some of the economies of scale problem by just embedding it in your applications and running it client-side. 2. GPT-4 is not the best every time: Cybertron-v2 (at 13B) did better than GPT-4 for writing and creative tasks - it's subjective but the fact that I thought so is insane. We're looking at a future where specific models are used for specific tasks, and any meaningfully complex AI application will be multi-model. 3. GPT-4 is still the best on average, and there's no real competition: If you're looking for a model to plug into every task and get reasonable output, use GPT-4. Maybe Claude-2, but you'll need more time prompt engineering. 4. Claude-instant is underrated: at 10-15x the speed of GPT, and way cheaper costs, it's performance is pretty close to GPT-4 and well above 3.5 in my tests. Speed and cost make a huge difference - it's becoming my go-to model for ancillary tasks (fixing, labelling, etc). 5. There are no reasons left to use gpt-3.5-turbo. It performed abysmally in my testing, Claude-instant is cheaper, GPT-4 was actually faster, and it's become pretty bad at following multiple instructions. 6. OSS models still have a deployment problem: I've tried nearly every provider (Replicate, Vertex, Modal), and the cost, cold boot, time to first token, and generation speed are all pretty far behind what you can get from the big providers. It's likely that none of them have the economies of scale the big guys do on one or two model flavors. When you can't saturate H200s on a single model, and are forced to serve multiple finetunes or run arbitrary code of off-the-shelf cloud offerings, you likely have huge inefficiencies that may never be surpassable. This really is the big problem for OSS models above 3B-7B. How do you deploy them? Even with custom finetunes, how do you justify the cost when deploying to customers - both in running the model, and in the devops time managing things yourself? When you can plug in and API key and get sub-second time to first token on GPT-4/Claude-instant, why do that? Imagine if you (or your provider) needed to spin up a separate VM for every edge function invocation on Vercel. This doesn't apply to the 3-7B models I think. There are huge privacy, speed and cost implications to having weights and models embedded in local apps (through the work of people like @vatsal_manot), for things like natural-language-to-X, semantic search, conversational interfaces, tool usage, etc. WDYT?