kinj28

I would like to imagine pacing frontier is in interest of both (anthropic and open AI) as it will enable them to enlarged their depreciation

I would like to imagine accounting inference revenue on trained model and the depreciation cost for training that specific model must already been capitalized + compute to serve would be a net positive margin business. Ongoing training must rather be for future models.

But again once future models arrive they would render older models useless, so the asset must be depreciating really fast.

Would love someone to throw light on revenue and cost recognition at the unit level for this.

My father has gastric melonama and pretty much last few days left — does someone with knowledge know if this vaccine can be of any help? Or any registry available?

I cold emailed a researcher in Japan about gut bacteria e. Americana for cancer treatment and registering my dad for clinical trials as and when they happen. And hell yea - professor replied :)

So yes — just do it!

But isn’t the case that faster you fail you get to the goal post better? Terminally ill trial vs non terminally ill can be differentiated somehow while measuring efficacy and other post trial parameters ?

This seems very promising

This is helpful.

I would think everything boils down to source of funding and their narrative should get pushed!

I am afraid — if Chinese models go mainstream it has a clear way of pushing its narrative way beyond its otherwise borders. More like a Trojan horse it is for the Chinese.

Here is a quick example of how Chinese deepseeks agent works kn its underlying model) when asked a tough question

https://x.com/jinen83/status/2079406993979383902?s=46&t=D7hQ...

Maybe a tangent point but quality of their open weights is a directly proportional to what’s gone in while training. IMHO Chinese models must have gotten heavily biased “official Chinese view” in most topics of how it sees the world. So for AGI purposes — yes a challenge would be win in the west for these open weights

I tried DeepSeek agent to get answers from Chinese models on some tough questions regarding Chinese govt and it refused. I am very keen to go a level deep and host the model and see what it really gives an answer

https://x.com/jinen83/status/2079406993979383902?s=46&t=D7hQ...

I am using it as one the agent that is automating LinkedIn outreach by running a bash script & using ai wherever it needs some decision like finding first name or what message to write, etc.

Makes sense. But with plain English you can build similar small single purpose app and someone else hosts it. No need to learn how to make low code work and 10x faster. IMHO plain English will triumph!

DronaHQ.com

The day they are available as a platform to teach them skills and make them do things like we need it!

Some random thoughts —

why language would evolve ? Let’s say to make it easier and better ? And if such a case then wouldn’t that be applicable to all languages? If yes then I am a native kutchi speaker and it just a dialect. How would its history of change could even be found? But I do speak other languages like Gujarati and Hindi and I wonder if there was any evolution if those languages which have a

As I read the article — I was curious if there are any language museums. If any would love to visit.

They do give an option to upload statements. I don’t know if that leads them to achieve claw back. Am curious if they can do this?

I am the founder of Dronahq - bootstrapped Low code platform. I fully concur with how AI is eating low code platforms for breakfast.

However underlying principles haven’t changed.

-Engineering bandwidth is minimally available for Internal tools. -Enterprise controls/guardrails are important needs - bringing in data to your app is a must have - maintaining code vs low code apps — low code has been a lot easier

In a conversation with a CTO at VC fund - he predicts that 4-6 quarters and you shall see demand back to peak in low code segment!

In a customer conversation— customer made one tool with cursor and he was very successful but by the time he started adding features for 2.0 everything started breaking and he wanted to move back to lowcode.

As a low code vendor- we just added internal tool building agent that underneath writes react code and leverages the other core capability of the platform thereby giving users best of both the worlds.

But surely interesting times ahead for the category. Let’s see if it survives or dies!

My personal take— it will survive and converge with agentic ai!

My cofounders mom had got Alzheimer’s at rather younger age of 60 something. It grew extremely rapid and within 2 years she grew into a vegetative state.

Given how I have seen different patients and varying pace of dementia growth - I somehow think Alzheimer’s must have multiple sub families & may appear same but must be behaving differently from their onset.

Somehow I don’t see it and too lazy to inspect element

What I meant was there are collection of genes responsible for error correction. If there is a failure in duplication then these genes have not done their job.

Thought experiment, again as a layman, was to see if these genes responsible for error correction at the base level can be fixed or bolstered and that will act like a cancer vaccine. But looks like from other comments that this is even more harder!

Nivumolab was the drug administered in adjuvant setting. Maybe you are right that 400k patient with decent efficacy - however pegged chances are about 70-80% and not 100. So my point is can there be a better test bench to try and inch closer to a better efficacy?

For example - if hela cells can be used for trials — can there be the cultured tissue be used instead of mice as day 1?

Also curious — how did the scientist decide on using a specific cell/protein to be used for checking if this is producing results. Is it a hunch or science ?

In my dad’s case- he had gastric melonama. We surgically removed it and as consolidation We administered pd-L1 Immune checkpoint inhibitor. Melonama recurred again in 6 months time. This time in esophagus.

As an engineer I think all drugs tested and efficacies studied are on statistically not so significant data points. Given the permutations and combinations far exceed the clinical trials available and hence everything post clinical trial is also just an extended trial.

Wonder How to fix this? I am assuming heLa cells etc are also not the right test setup to have better test results.

I have been working with my dad on his cancer treatment since last year. My interest in the topic has only peaked ever since.

(Disclaimer- I am an engineer and not a microbiologist/doctor)

Mutations and wrong copying of genome happens all the time in the body and some enzyme has the job of correcting the mutated genes so it doesn’t get into the system. Level 2 defence is T cells killing it as identified as foreign body.

Thing that baffles me is that I see most work happening to eliminate tumor. To me it sounds a tough problem given the permutation and combination of mutation— roughly few trillions.

But I was curious if there is working happening on L1 defence — fixing the enzyme that fixes the wrong copy paste mechanism. Or making the enzyme get more efficient and powerful. Is that line of thought even valid?

I run a low code platform for building internal tools & software. One of my prospect about to sign a contract came back telling me that his CTO has asked him to check vibe code tools and build a few internal tools with them. They are a large series D/E company and have over 250 internal tools built on retool (a service that they are migrating from). CTO is puzzled & is thinking if does he even need a platform to build & manage internal tools.

On other hand -- another customer of mine built a few internal tools with vibe code (& yes he does have subscription to my low code service) but then when newer requests came for upgrade thats where his vibe coded app started acting up. His candid feedback was -- for internal tools vibe code doesnt work.

As a service provider for low code --> we are now providing full fledged vibe code tooling on top. While I dont know how customers who do not wish to code and just have the software will be able to upkeep these softwares without needing professionals.

ChatGPT tells me inventwood and zinnia from MIT are already at some stage in the lifecycle.

This surely seems like a game changer and won’t need much of deforestation at some point.

Can we get lab made wood then ?

I wonder how authentic the hat must be after restoration? And how exactly restoration is done ? It seems restoration had to be funded so must be some elaborate process.

Fixed it!

That’s good to know !