dean |

@depinnomad @FliteGrid mixing in a bit of UGC vibe, nice!

Seriously loving how creative people got with their @FliteGrid posts this month 🚀

It was fun while it lasted @Alibaba_Qwen

AI models are trained on roughly 20 trillion tokens, or every public text on the internet. and it's still not enough. The next frontier isn't more text, It's the physical world, and 375ai is already capturing it.

@nishancodes @tehface @RoundtableSpace I appreciate it! I checked through canirun.ai and it seems the performance for E4B and Qwen 3 8B are the same for me. If Qwen is significantly better and has basically 2x params, is there any tradeoffs? Why would an 8B and a 4B be almost the same re performance

Sure, Since it's a weaker model, it's tool selection capabilities are weak. It would often choose the wrong tool. In weaker SLMs the thing I've often seen is that the json output formatting breaks. multistep reasoning is basically crap most of the times. For your case, the model has to read the mail, summarise it and send reply. In SLMs this process fails often. If your usecase is something simple and occasional, you can go with this, otherwise go with Qwen 3 8B. It's heavily optimized for tool calling

GEMMA 4 CAN NOW RUN DIRECTLY ON YOUR PHONE WITH GOOGLE’S OFFICIAL APP. It’s fully open source, multimodal, works offline, and already runs on both iPhone and Android.

@nishancodes @tehface @RoundtableSpace Can you elaborate why the 4B is highly unreliable for tool calling? I haven’t dove into it yet that deep so I have yet to test and debug any errors/limits I encounter. How big of a gap do you feel when using Qwen 3 8B v Gemma 4B? Smarter? I’m new to this so I appreciate it!

So for making it read emails and play music/ videos, you need to make the model call tools right? 2B and 4B is basic for toolcall and highly unreliable. Qwen 3 8B is the good choice for your usecase. You have to write single and multistep tools for your needs. Natively the LLMs don't have this capability

@nishancodes @tehface @RoundtableSpace I’d compare it to the first iterations of Alexa, but locally. Configuring my jarvis to read emails, play spotify/youtube directly Small small QoL improvement but I think we’re just getting started. How do you feel about Mistral 7B / Qwen 3 8B?

@cryptdean @tehface @RoundtableSpace Yes, the 26 and 31b models are good for people with high end GPUs. The lower models are good but not usable for coding or reasoning tasks. It can be used for casual conversations

@nishancodes @tehface @RoundtableSpace Building a simple JARVIS with E4B currently and I can say it’s decent! imo it’s a great leap for those who have higher-end PCs and don’t want to rely on cloud computing due to privacy or costs

@tehface @RoundtableSpace Great. Which model are you running? 4B?

the world yearns for Filipino culture

It’s exciting to think that real-world data is important for JEPA and Spatial Intelligence

@MaziyarPanahi Which model were you using specifically?

Gemma 4 just dropped. I had it captioning video in real-time within an hour. Running locally on a MacBook. No cloud. No API. Real-time scene understanding. Oh and SAM3 is segmenting every object in the same frame. Same laptop.

@MilksandMatcha Never tried Codex. Do you guys also hit limits 15 minutes into the session?

Giving away 5 Codex Pro plans Each person will get 3 months of free Codex Pro (highest tier). Winners will be selected from comments in 48 hours, comment below why you want it.

Are we good....? @GeminiApp

"claude usage limit reached. your limit will reset at 7pm"

@wingbits Congrats guys! the future is orange🟠

Was it not pretty clear from the beginning? As much as I love the convenience of being able to drive around when I want to, it’s pretty clear that *we* are the traffic. Let’s redirect this energy towards demanding a better public transportation system for the entire country. I’ll be more than willing (and happy) to integrate commuting again when Mt. Kamuning, Mt. Shaw, and things like that have been improved for good.

Even the darkest clouds can have a silver lining. Because it turns out the most effective way to reduce traffic wasn’t a policy or a new flyover, but an oil crisis. Prices went up, optional trips went down, and the road breathed. And it also raises a question nobody wants to answer: were we ever honest about who was clogging the roads in the first place? When fuel was cheap, the threshold for “I’ll just drive” dropped to almost nothing. Optional trips. Convenience runs. Vehicles that had no business being on the road except that the cost of being there was affordable. When the price corrected, that optional travel quietly disappeared. Not all of it. But enough to move the number 8 points. Now I know how this sounds. I’m not celebrating the price hikes. People are hurting. The tricycle driver. The jeepney operator. The TNVS guy moving forty passengers a day. The family in Cavite driving two hours each way to keep a job they can’t afford to lose. I’m not dismissing any of that. That’s the dark cloud. The silver lining is there’s also math nobody wants to do. If fuel costs more but you arrive in less time, the total cost of that trip hasn’t shifted as dramatically as your receipt suggests. Because not only does traffic burn more fuel unnecessarily, there’s also your time, if you value it at all. The pump price is real, and yes, we need to cut the excise tax and VAT during this crisis because it affects the price of everything. It just isn’t the whole equation. It’s not unlike a toll road argument. We choose that over the service road for the same reason. Because let’s face it, eight percent doesn’t sound like much. Until you’ve sat on EDSA.

@SageDatuin @VALO2ASIA @heyimquack @paperthinhere @blackenBluePH @RiotJaker been saying since 2021 #MastersManila goes hard

@VALO2ASIA @heyimquack @paperthinhere @blackenBluePH @RiotJaker I would love to see a Masters Manila one day 😭😭😭😭😭

VCT Pacific Finals in Manila 🇵🇭, Mumbai 🇮🇳, and Sydney 🇦🇺? WE WILL BE THERE Watch #YAP2ASIA EP. 20 on YouTube, Spotify, and B站 @heyimquack @paperthinhere @blackenBluePH @RiotJaker

🤦🏻

Number 3 is where it’s at It should be something people actually use, and they can opt in to contribute and earn rewards. Utility-first principle people never went on facebook to earn money. they knew their data was being taken for free and they didn’t care. they loved the product manufacture dashcams PRIMARILY for dashcam use, then let users opt into something like @ROVR_Network integration the biggest DePIN project to come won’t even be explicitly labelled as DePIN

Just spoke with a large-scale DePIN deployer who stepped away from the industry. A few things we agreed on: - Considering price volatility, he considers token incentives a disincentive at this point and would prefer to deploy for projects that pay at least partially in stables. - The future of DePIN is private equity finding web2 companies that they can convert to the DePIN model. - We'll see more and more hardware that serves some purpose as a consumer product, and it will be marketed to consumers instead of investors despite being a DePIN device.

@suburbancrypto @375ai_ @0xQuez @DePINConnection @harrydewhirst burby is EATing good

Was hungry for more $EAT , had to go to Target and get those sweet sweet tokens thanks to @375ai_ @0xQuez @DePINConnection @harrydewhirst

@katexbt (:

And another one bites the dust. Here we go with Grab monopoly in South East Asia…