sabyasm

@nbaschez Hello! Yes please!

Do you spend a lot of time reviewing markdown docs written by AI? Wish it were a better experience? Say hi if you wanna try a new (free, open source) thing

@tankots @WisprFlow Wispr Flow

We offered 5 people a Porsche 911 GT3 RS if they could get @WisprFlow to make a mistake It's the fastest and most accurate AI voice dictation app that's 3x more accurate than ChatGPT, Claude, or Siri. Today, we’re finally launching on Android. Download now: play.google.com/store/apps/det… As a part of the launch, we’re giving away 6 months of Wispr Flow Pro for free. Like, retweet and comment ‘Wispr Flow’ to get it. Enjoy. — Written with Wispr Flow

shady move by X. for those without context: >Selo created a mega-viral and profitable app that ranked users on X >interviewed at X, they ask him to explain the architecture of the app in extreme detail >interviewers seem super impressed, app is far from trivial and users really love it >great, hard work and product instinct is paying off, he’s about to get a high-paying job at X >not quite >X sends him a legal notice like hey take it down or we sue you >also btw you’re not getting the job >weeks later X announces a feature to rank users on X i don’t like it.

@NanoBanana Generate an image at location 26.29821252110648, 73.01867569451082 with a bunch robots in ironman suit marching, at the sunrise

@NanoBanana Generate an image 26.29821252110648, 73.01867569451082 with a bunch robots in ironman suit marching, at the sunrise

@RihardJarc @grok @AskPerplexity what's the source?

A Former $MSFT employee who worked with OpenAI on the $MSFT side shares some really interesting views on the relationship between OAI and $MSFT and why he thinks $GOOGL Cloud is the next wave: 1. In his view, there is a plateau as of GPT 4.5. forward in terms of training. He thinks that GPT-5 is GPT 4.5. with a router on top of it. »That thing was failing, and they had to go to essentially a router system«. 2. One of his goals at $MSFT was to »babysit« the OpenAI IP. If OAI declared AGI tomorrow, $MSFT should be able to code up ChatGPT. He thinks that $MSFT has a good view into the algorithms and computational primitives that sit there. 3. He thinks that OpenAI released open-source, because they are trying to flood the market with cheap and very good AI, so everybody has to lower their costs (Anthropic and others). With that, the $MSFT exclusive IP license also becomes less valuable. 4. He explains that $MSFT is not married to OpenAI and that Satya is basically saying internally to use whatever model is best just to get value to the customer. He believes that Satya realized the moat with OpenAI was shrinking and would continue to shrink. He thinks that the trend of people moving workloads to Azure because of the OpenAI brand, and with it, the latest and greatest that comes with it, is only temporary. He also believes that $MSFT thinks that ChatGPT Enterprise is a direct competitor to what $MSFT is trying to do. 5. He thinks that both GPT5 and GPT6 will have incremental improvements for you as a customer and that the additional value will only be there if you are working with an academic institution doing research. He believes that LLMs are commoditizing. 6. He mentions that he wouldn't sleep on $GOOGL Cloud as he believes $GOOGL will have a serious moat and could attract a lot of customers to their cloud going forward, as they have the intellectual capital and research arm together with things like Veo three. He thinks $GOOGL Cloud is the next wave. found on @AlphaSenseInc

Let me tell you a dirty secret about a lot of execs: They're extremely smart. And they haven't had to do their own work for years. Look inside any mid->large size company and you'll find VP+ executives that were promoted fast and furious in their early career because they're smart, hard working, make good decisions, have good taste, and can manage up down and sideways well. And as they become more senior, they start to earn the "you're too important to [X]" executive scaffold: - EAs for admin/scheduling/todos - Chief of Staff to keep their directs on track - Sr. leaders working under them eager for opportunity, so take on projects, presentations, meetings, etc. They're still smart, and they're still hard working, and they still make good decisions, so they tend to orchestrate and use these tools at their disposal quite well, choosing what gets done by whom. But - they show up to board meetings with decks made by their team - they show up to sales meetings with prep docs done by the sales person - they share insights generated by some data team - they +@[ea] to schedule every meeting - someone reads & responds to their emails And their job becomes - charm customers - charm candidates - charm the team - charm the board - charm the market - have good ideas (for someone else to do) And before someone shouts "this just optimizes for people who are highly political!" I must emphasize: these people are still usually wicked smart, they're usually extremely charming, and they work really hard (earliest on, latest off.) They hang in the forest, not a tree. Their experience saves your butt once or twice. They just don't have to put their hands on a keyboard and do the things. Sometimes they *can't* put their hands on a keyboard and do the things, because they're in endless meetings and on endless trips to do the charming/idea things. But after awhile of this, their "doing the things" muscle atrophies. Your CMO can't write compelling copy. Your CPO doesn't look at designs anymore. Your CRO can't login to demo. Your VPE doesn't have the latest local env setup. And over time, as an exec, your ability to be wicked smart degrades with your distance from the work, especially when things like AI come and smack your team in the face. This is all to say, I have two warnings for you if you've made it this far: EXECS: Do not stop doing the things. Take on projects, write your docs, do your own damn analysis, and don't stop touching the work. Fix your calendar so you're not just bopping meeting to meeting, and use that big brain of yours do actually build something. TEAMS: Consider yourself lucky if you have a leadership team that can still/wants to do the things. Trust: you'd rather have a micromanagey CEO who drops suggested edits in your doc than a manager-class exec team that doesn't even know your doc exists. Your company will be better for the work, the specificity, the care that comes with a doing-the-things leadership team, than a organizes-the-work leadership team. And for all of you: AI will generate a little microcosm of the dynamic above, but for IC work. I love AI, but still think it's important to exercise the muscle daily of writing, coding, reading, speaking, thinking. All unused skills will atrophy. Make sure you get stronger, not weaker, with these new tools. Ok back to doing things 😎

@BillyM2k Can you make it in p5.js @grok

when a ball escapes, 2 balls appear inside

@mukeshsoni The's the dream - vanna is actually pretty good.

@sabyasm But what if you build a great text to sql solution which is built on top of a custom model which is super small and can be deployed on prem?

The text to sql space is flooded github.com/vanna-ai/vanna github.com/Canner/WrenAI github.com/mindsdb/mindsdb julius.ai Key differentiators would be 1. products ability to connect to various types of databases and 2. latency of user query to result 3. Accuracy

1 & 2 are somewhat solvable. 3. is extremely difficult to do unless the AI has access to business glossary, metadata, sql logs and execution plans (what has worked) + a human in the loop interfacing to course correct. This is why I feel data platforms like snow, databricks etc. has a strong moat than opensource/small players.

@mukeshsoni + Snowflake [youtube.com/watch?v=va-l7s…] + Databricks AI/BI Genie

@nutlope Love it. How do you design your front end and what do you use for video recording? Congrats on the whisper release - love the idea and execution

Here's my exact process for building viral AI apps. In this talk, I go through: ◆ The architecture of a typical AI app ◆ My entire process of building, from ideating to launch ◆ My top 7 tips for people building apps

@mattyp - could we automagically include this in @replit?

@AJA_Cortes Veterinarian business: PE firms are cutting jobs, and both employees and vets are exhausted from constantly pushing vaccines, medications, and premium pet food. Honestly, it’s gotten so bad in Canada that sometimes it feels borderline illegal.

What's a formally good business that you experienced be ruined when bought out by private equity? I want to hear people's stories

@mukeshsoni Ok - you might want to try this repo. Seems like they have spent significant time on the the video/voiceover sync issue as well #L75" target="_blank" rel="nofollow noopener">github.com/google-gemini/…

@sabyasm I have come across multiple cases where there is information only inside code in some repository. No one wrote a blog post or made a video explaining how to do it.

Is it possible to generate a technical video (like a person taking and live coding) by making the AI look at the commits of a git branch?

@mukeshsoni I am tempted to try but who do you think will be interested in that video?

I think so, you can use CUA or even something like open ai's operator or similar open-source offering to do the typing and browser automation meanwhile the same ai that's calling the browser automation generates the human voiceovers transcript (or real-time voice generation through elevenlabs) as the flow goes.

@mukeshsoni The newer version of the product does search sometimes, but switches to shortcuts depending on the load, etc., I believe. It proves the point of your Q about how much you’re willing to spend to get a better result. future is here but not evenly distributed

@sabyasm I thought the way i asked about books and articles, Chatgpt would automatically do a web search and embed links from that search. Strange that it instead made links up.

Asked Chatgpt a question on raw image processing. I needed names of books and blog posts which go into details of raw image processing. It gave me names of books and a lot of links to articles. Most of them were made up! Whole book and author names. Urls. All mde up!

For anything that needs real world data - vanilla LLMs are pretty unusable. By design they hallucinates 100% of time - but most of the time that's useful. Like brainstorming etc. For this type of use cases a, I use deep research or grounding the information with google search. My default "chatgpt" interface now is google's aistudio with grounded with google search option enabled.

Blog post urls which never existed. I thought the server for the blog post urls might no longer be there and checked the urls in wayback machine. And i found out those links never existed. The domain itself never existed! 😳