makingkaiser

🇸🇬🇺🇸 | Inference @FireworksAI_HQ

[makingkaiser @kaisereatsbread]

@benln Just curious, why B10 and not Fireworks?

[Fireworks @FireworksAI_HQ]

We ran Kimi K3 against Fable on ~1,000 agentic tasks, expecting a catch-up story. We got a specialization story instead.

@kimi_moonshot's K3 outperformed on security, crypto, and long terminal loops. Fable beat on multi-lang + web/data viz. Per-task routing hits 93% accuracy, above BOTH models, at up to 50x lower cost than Fable on long loops.

The part nobody's pricing in yet: the router sends 72-96% of traffic to K3. The frontier model becomes the fallback rather than the default.

Kimi K3, coming to Fireworks July 27.

[Fireworks @FireworksAI_HQ]

Our NYC team is growing!

Just in case you needed another reason to visit https://t.co/9TyRcuswYh

Come build with us.

[makingkaiser @kaisereatsbread]

@brianchew PLEASE

[makingkaiser @kaisereatsbread]

@brianchew hold on to your papers, fellow scholars 🛑

[makingkaiser @kaisereatsbread]

@brianchew haven't seen one yet

[Lee Robinson @leerob]

You can now try GLM 5.2 in Cursor!

Excited to see more useful open models, thank you to Fireworks for partnering here. Results from our evals ↓

[makingkaiser @kaisereatsbread]

@DennisonBertram @Zai_org @baseten @FireworksAI_HQ Hey Dennison, would love to hear more about your experience on Fireworks. Shoot me a DM!

[makingkaiser @kaisereatsbread]

wow. Zero kld!

[Fireworks @FireworksAI_HQ]

The hard part of reinforcement learning on a frontier model is the infrastructure that keeps training and inference numerically identical: zero KLD, end to end.

We've solved this challenge, and are now offering it as a managed service, starting with GLM 5.2.

[makingkaiser @kaisereatsbread]

@ivanleomk @FireworksAI_HQ Hey Ivan, feel free to DM me!

[makingkaiser @kaisereatsbread]

@spyced @jeremyphoward @FireworksAI_HQ historically they always release a fast mode for most models. should be the case going forwards

[Jeremy Howard @jeremyphoward]

Wow.

@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.

It responds with nuance and judgement, & handles long context VERY well.

I've never experienced an open weights model like this before.

[makingkaiser @kaisereatsbread]

@jeremyphoward @Zai_org How much of this do you think comes from distillation?

[makingkaiser @kaisereatsbread]

@nrehiew_ Probably just invent a new benchmark on safety, under the guide of pretending to be a compliant model

[橙子🫪 @tangeorange]

中文的魅力

[makingkaiser @kaisereatsbread]

why isn't there a webox cli

[hero thousandfaces @1thousandfaces_]

Insane how you could be in the coolest place ever but then you get back to California and all of the colors are like 1.5x more saturated and the light is slightly brighter. Absurd place

[ethn @ethanyish]

🇸🇬There’s a take I keep seeing that Singapore doesn’t produce founders. The standard reasons given are that LKY was the ultimate founder so no one else has to be, that the system is too homogenous, and that the market is too small and fragmented (South East Asia). I’ve read and produced versions of this argument for years. I’m not going to re-litigate it today.

I’d rather share what I actually see working. Singapore has produced tech founders. Razer. Creative. Sea. Peng Ong co-founded https://t.co/GO61fvQzuQ out of here. It’s not a long list and no one pretends otherwise, but the list exists, which means there's precedent.

In non Singaporean fashion, I want the discourse to shift from the 1,000 reasons singapore can’t produce founders, to the reasons singapore could produce great founders and how to make that work.

If @ycombinator managed to crack the code 21 years ago to generate 10,000x times the number of founders and changed the mindset of the entire silicon valley, I do know that Singapore and the broader ecosystem will have its own way of achieving some outcome here. YC normalized failure as a young founder imo, it wasn't as hot and sexy to start your own company then.

There are green shoots - a new cohort of Singaporeans abroad quietly building real companies. Jeffrey Tiong from @Patsnap. Desmond Lim from @workstream_us. Thomas Li from @Daloopa1. It takes 10 plus years for these success stories to grow and reach the mainstream in Singapore.

How can singapore founders combine the best and brightest and most hardworking Singaporeans with the abundance of global opportunities available in this new age of ai disruption? A lot of this having the mindset to sell globally from day 0. And of course one of the biggest markets in the world is USA. Is this easier said than done, yes. but is this definitely doable? yes.

At @AlliumLabs, we have a Singaporean core founding team (@lchylch @potaypaul @gohchaa) incorporated in US. We benefited greatly from having a Singapore office from nearly day one. HQ in the US, usually New York or San Francisco, because that’s where the deepest pool of capital and the largest customers actually sit. I do think combining the best of both worlds is a viable path forward and a template that one can expand and scale on.

We actually ran this pros & cons (in typical singaporean pragmatic engineer fashion)
1. Are there critical and core parts of our work embarassingly parallel? Output of one team is not blocked by the output of another.
2. Did we get 24/7 coverage SLAs for our customers because "the sun never set" on Allium? Yes. This allowed us to service the biggest institutions in the world at a much younger stage in a leaner fashion.
3. Can we get access to top tier problem solvers who have great work ethic? YES.

We mitigate the timezone piece by having the Singapore team log in ~10pm-11pm SGT Monday to Thursday for more synchronous communications with the US team. Is this easy? Not really, is it possible and a sustainable template going forward? 100%.

Allium is built this way. We’re headquartered in New York. We opened the Singapore office in year one. The founding team is Singaporean. Our customers are global. Both offices do real work, and neither is a satellite of the other. The reason this works is that you get both sides. Singapore gives you strong operators, real financial infrastructure, access to Asian markets in one flight, and an institutional reputation that’s hard to replicate elsewhere. And of course the ability to hire the best and the brightest in the broader region. The US gives you customer density, deeper capital, and the kind of ambient pressure that forces you to think bigger than you would at home.

You don’t have to pick, you can have both.

If you’re a Singaporean sitting on an idea or a hope to do some great on the global stage one day, that’s the thing I want more people to hear. You also don’t have to wait and say it's hard. Of course it's going to be hard. Just set something up and keep going.

Here's what I am happy to do with fellow sporeans
1. For Singaporeans who do want to build their own company one day, I am happy to set some time to share my own personal journey to cross the mindset chasm.

2. We are expanding our office in Singapore of course (just got a new lease in Chinatown), and imo this is the best way for a Singaporean to get exposed to the global tech world while still being in Singapore and being "not just a satelite" office. It's an MBA for startup. If you're interested in opportunities too, let me know

I have personally cared about this topic for more than 20 years (reading sim wong hoo's, creative's founder 1999 book) as a kid, and have spent most of my life thinking about it and what can be done and how it will be done.

[makingkaiser @kaisereatsbread]

an ad for an exhibition on singapore's public transport system

[wassieloyer @wassielawyer]

Yep this is the correct take.

Running a startup in Singapore is so negative EV that you select for retards who failed out of system, high functioning autists that refuse their place in the system and wealthy heirs trying to make their name outside the system.

The retard class is almost uninvestable, except for those who pivoted into the shadow economy (nightclubs, ktvs, gambling etc) in their teens and still maintain a code of brotherhood but good luck getting access to these people or convincing them to let you ape their KTV / Johor Vape Run. Avoid the system dropouts that pivoted to insurance sales like the plague (they basically sell financial scams to the elderly) and they will rug in the blink of an eye too. You might find a diamond in the rough that previously worked at a local tech company like Grab, Shoppee etc but it’s like finding a needle in a haystack bc it’s hard to tell them apart from a retard that got fired and will be pivoting into insurance if the startup LARP fails.

The high functioning autists are more investable but there are two main problems. To clarify - these are kids preordained by the Singapore system for high finance, bigtech, biglaw, medicine and other “high value” professions alongside a 2 million dollar mortgage but have somehow rejected their destiny.

The first problem is discoverability. The first class of autists fuck off to the West Coast, London or godforbid Berlin for college where they embrace and explore all their quirks that made them feel out of place in Singapore. The ones who went to Berlin are lost forever. The ones that went to California / London might, after a few years of self-discovery often involving sex, drugs and rock and roll financed by autistic superpowers or their parents, return to Singapore after either a family tragedy, the false promise of a good Asian wife / husband or their Western tax bill. You might be able to invest in them then as they seek funding for some grand moral mission to embrace or reject capitalism in their autistic niche area.

But the second class of autists that are investable don’t leave Singapore at all. You will find them in STEM, geeking out in the laboratories and computer science departments of universities, possibly in A Star. They speak a maximum of five sentences a day and are largely incapable of maintaining extended eye contact. But behind those bespectacled eyes in the acne scarred face is the potential engine of a billion dollar empire. Or a Chinese spy. Too bad you don’t speak autist.

The second problem is economic alignment. The autist does not understand money the way you do, if they did, they would be on half a mil+ comp packages within the system. You think just because you found them and funded them they work for your bags? No sir - they work for the mission. And you better pray that mission has the side effect of increased enterprise value or a path to liquidity. If not - thank you for your donation we will be applying our next gen quantum computing capabilities to solve poverty in sub-Saharan Africa.

Finally, the wealthy heirs. Success in the system means nothing because they will never emerge out of their parents’ shadow. Their grandfathers name is on your university lecture theatre.

Desperate to make a name for themselves, they will reject the family mandate to manage the real estate portfolio in order to found a startup. They might even take your venture cheque even though your fund is smaller than his inheritance bc an external fund provides “validation”. You will vie for allocation alongside people who want an intro to their dad.

Pros - they have near infinite funding and networks.

Downside - they don’t care about money and might be retarded. Good luck getting an exit.