Thomas Morselt

@bramk Processed review in one shot? Any learnings on what made your app pass?

Just got my first iOS app approved in the App Store! 🎉

@pixel_updates Is it available in Europe? I setup my sports teams but haven't seen the teams when they were playing

Live sports updates in At a Glance widget is pretty cool.

🏴󠁧󠁢󠁥󠁮󠁧󠁿 Premier League Predictions (Weekend) Share yours below ⬇️

@Quintin24 Mooie dump vannacht 👍. Was ook geen overbodige luxe

Zoon Mats houdt ons goed op de hoogte over de condities ..... 🇦🇹

@Quintin24 Laat je nog wat sneeuw over voor volgende week?

Kaiserwetter

@ai_for_success On Ultra or on Pro account?

Official: Google has added Opal to the Gemini web app as an experimental Gem, letting users build reusable AI powered mini apps from the Gems manager.

@OfficialLoganK Yes, please for the early access.

Reply here or DM me :) will add folks in as much as we can

Big upgrade to vibe coding in @GoogleAIStudio lands in Jan, but if you want to test early… 👇🏻

@kimmonismus I mean, most people don't even get to use deep research. That is the main target audience of Microsoft and Copilot.

Microsoft is always late to the party - and somehow releases products less useful than the competition.

@joshwoodward Rightfully so, this growth. Hockeystick is ready for 3.0. But what is it with AI teams and graphs?

🚀 Let's go!

@darnocks @chamath Firebase Studio does a nice job when adding in cloud hosting on GCP and firebase database with authentication, writing and reading data from a database. But you are hooked on GCP

i totally agree, thing is we've seen this exact pattern with hundreds of devs. that wall you're hitting? it's because ai-generated code scores 2/10 when reviewed by senior devs. works for demos, explodes with complexity. we tracked it: only 15% of ai code survives to production unchanged. the other 85% gets rewritten when you add that "necessary complexity" - the hover states, transitions, all those tiny details. one founder spent 3 weeks just trying to add payments to his lovable app. authentication was in the frontend. passwords in plaintext. total chaos under the hood. the problem isn't the ai - it's that it has no architectural context. it's been trained on millions of examples of bad code. so when complexity increases, it doesn't know WHERE to put things. we built builderbox to fix this. instead of another boilerplate, we give your ai perfect architectural instructions. database patterns, component structure, where business logic belongs. suddenly your ai writes 8/10 code from the start. that founder stuck for 3 weeks? fixed payments in 45 minutes. you keep the cursor workflow you like - narrow context, approve edits - but now the ai actually knows what good architecture looks like. the vibe zone expands when your ai has a map.

It’s unpopular to say, but it’s true: You can’t vibe anything useful rn. You can barely vibe a working product. And there are still virtually zero examples of anything even moderately useful/successful that was vibed - especially considering how much money has been spent so far on LLM calls to try and do so. This won’t be true forever but is true today. As with other tech cycles, a trough of disillusionment may soon come as folks get frustrated and give up. In the meantime, trying a more team oriented approach to building may result in better outcomes. Our Software Factory was extracted from our work making useful software for large, demanding enterprises. It’s built for teams to be able to build production software together. Alpha users rolling through now. Beta in Sep and GA in October. 8090.ai/waitlist

@koltregaskes Looks very do-able

@thomasmorselt Yes, seems to undercut their various levels of models so this eas clearly the target.

I'm disappointed that GPT-5 is text- and image-only, with no audio or video input/output. The model page mentions image generation via tools, so must be 4o image. I wasn't expecting it but hoped for image v2. I'll stick with Gemini 2.5 Pro for video and audio uploads. The October 2024 training cutoff suggests pre-training was long ago, with performance gains from post-training. This is fine, as models have web access, so a recent cutoff would only be a bonus. Overall, it feels like a GPT-"4.5" - more "cool" than "wow." It's a decent improvement, with better reliability and instruction following, addressing key bottlenecks Routing issues will be resolved in the coming days, weeks, or months I'm sure, but the Thinking toggle helps and apparently "think hard" prompting works. Pricing is competitive, undercutting Gemini models in respective variants, and it's fast (I like the "quick answer" feature). They're likely compromising to get it out to the masses and cutting costs. And the routing problems were always going to be the big problem; give them time, and I'm sure they'll fix it. Overall, though, this is a good model, but it was never going to reach everyone's crazy expectations. Thank you @openai, but we need a GPT-5o now. ;-)

@koltregaskes Have it on Plus account in 🇳🇱. First thoughts; Unlocks a lot of very specific use cases, but needs even more specific instructions for reliable outcomes.

ChatGPT Agent now rolling out to OpenAI's Team plans it seems. Anyone who still doesn't have it? Anyone who does, what are your thoughts and use cases? All Pro users should have it, Plus started rolling out a couple of days ago.

@koltregaskes How did you install the browser before the invite

Oh yeah! ...though I have been able to download and install the browser already. I'm wondering if this unlocks more features?

@Quintin24 Ik hoop veel, want het zal iets neutraler zijn dan de social media feeds

Hoeveel mensen zouden voor de volgende verkiezingen met AI hun “eigen” stemwijzer maken?

@kimmonismus Is that a 4 million context window (output)?

New Gemini 2.5-Pro version about to be released soon. Big jump on Aider polyglot 76,5% --> 86,2% surpassing even OpenAIs o3 (81,3%)

@MatthewBerman You won't regret it and probably just need 2 hours to get used to it

Going to the Google Store after this and buying the Pixel 9 Pro XL.

@jowyang Are the sides from your keynote available for sharing? @jowyang

My final slide from yesterday's keynote at AI Realized conference on where I think the AI agents space is going in regards to enterprises: Yes, your future customer/partner/employee will be an AI Agent.

@btibor91 o3 DR works very well with Excel/CSV sheets. Can create comprehensive reports because o3 has Python functionality. Had it analyse and create a report on one single Excel sheet for 26 minutes with very good results.

Summary of my findings after comparing more Deep Research reports (these are my personal opinions based on my own tests and experience) - No perfect solution yet, since all current deep research tools have limitations and make occasional errors, so you need to verify the outputs, because relying on any single solution could lead to incomplete or biased research results - Combined approach works best in my opinion, because the best way is to use multiple tools in parallel, then validate, verify, compare and combine all findings to get what you need, which saves a lot of time compared to traditional searching and browsing - 1. ChatGPT Deep research (o3) is probably the best one right now, because it goes into a lot of detail with very powerful search capabilities, offering exceptional depth, best-in-class search, most comprehensive for critical research, and consistently provides detailed reports with solid references - 2. Gemini Deep Research with Google search expertise delivers quality results, and is available even for free accounts despite high capability, but the writing style is a bit verbose like a scientific wall of text, and it performs strong in technical topics - 3. Grok DeeperSearch offers efficient, quick, focused research with a good balance of speed and accessibility, and while it's not as detailed as ChatGPT or Gemini, it's faster and performs well for time-sensitive research, making it great for initial exploration - 4. Claude Advanced research is only available for Max, Team and Enterprise accounts for now, and while it takes its time with sources, the reports are less detailed, with consistent but not outstanding performance - 5. Perplexity gives more surface-level analysis with limited depth on complex questions, and lacks the full capabilities of the others, making it not enough for detailed research - All tools already provide mostly up-to-date info, good relevance to research questions and decent structure in most reports, but the biggest differences are in depth, evidence quality and how transparent the methods are - Always verify claims across multiple platforms, check citation quality, use focused and detailed research prompts and compare outputs before making important decisions

@ai_for_success New memory being too real?

ChatGPT acting like angry girlfriend 🤪