Deepak Dhingra

@pranaykotas Keeping the model open serves them in multiple ways, they are attacking the Indian economy, with every new release, the value of software development declines, hurting Indian IT services while also disrupting the American SaaS market.

After today's Xi speech on AI, my falsifiable prediction is CPC will impose export controls on its frontier open-weight models before January 1, 2028. This could take the form of licensing requirements, delayed releases, parameter-count thresholds, or end-use restrictions.

Crazy ideas from the past DigiScents iSmell- A USB device from 2000 that synthesized smells from digital files, so websites and games could have scent. It raised serious money, signed real partners and became a permanent fixture on worst-inventions lists

TIL, In 1994 Bill Gates and Craig McCaw funded a plan for 840 satellites in low orbit delivering broadband internet to the whole planet . Launch costs made the math impossible, the constellation shrank on paper year after year and the company suspended work in 2002 after burning through a fortune. Starlink hy spaceX is Teledesic executed twenty five years later

@gregorykennedy I agree, my broader point is that for the industry to move forward, the best knob to turn will be the metrics people are judged on

@WuTangAI I suppose it depends on the kind of marketer one aspires to.

Most of what passes for startup marketing is just founders copying whatever Stripe did five years ago and wondering why it doesn't work in 2026.

TIL, vulgur language basically meant latin language spoken by common people.

Cluely is championing social media marketing on Instagram, their whole feed is like micro drama Big fan of whoever the ling long guy is, tried using the product unfortunately the onboarding of the product isnt that good and couldn't use the product.

Feels like every bullshit bio marker is basically a billion dollar startup space, create a bullshit signal capturing device and find bio marker maxxers they have a monthly budget to try every new fad.

claude for office work, for everything else chatgpt-

There have always been two kinds of malicious actors. Type 1 - People with bad intentions but limited intelligence or capability. Historically, the friction involved in causing serious harm was high, so they could act only in limited ways. Type 2 - Highly capable people with bad intentions. They were much rarer, but when they acted, they could create significant damage. AI changes both categories. Type 1 can now act on harmful intentions far more frequently because the barriers to capability have collapsed. Type 2, meanwhile, can operate at a global scale and becomes a genuine systemic threat which is one reason biological work is banned in Fable.

Before you insist that having access to AI is going to revolutionize terrorism please remember that dumb people are dumb and access to smart people is something they already have and do nothing useful with, because they are fucking dumb.

5.6 is just terrible at most tasks, the hype is all manufactured.

the most essential and underrated skill for an FDE is understanding the politics of the customer environment, if you go out there and start building with the wrong team then there is no way your product will get adopted.

Asking the model to follow a goal, or refer something before building have limited value, yes its good but when working in a session with 400k tokens already the context carry some many contradicting principles that it become impossible for the agent to follow your initial request, I keep going back to @trq212 talking about map is not territory, for an agent to actually follow the local session goal, the context needs to mapped out properly - it can't be a general follow the goal kind of prompts

Shankar mahadevan of fingers

@EricSpencer00 @gabriel1 I understand the concern but in reality an ai is less bias, makes you more comfortable when answering. Personally, am more comfortable with an ai judging me vs a human judging me.

@WuTangAI @gabriel1 personally I'm turned off by ai interviews and would rather answer a QA or some other filter questions I agree this filters the spam and pray, but what happens to people who don't apply bc of the added barrier of a one way video interview?

i especially hate CVs. what the fuck am i supposed to learn there? it's literally negative information bunch of high level abstract sentences. "leading a project" or "making something 40% more efficient" means nothing send deep insights that no one else could have or something

Nope, ai evaluates the candidates on questions that are essential for you while resume is aligned to candidate, You can skim through 30 candidates in 30 minutes. Half people applying to jobs dont even match the bare minimum criteria like visa, skill, experience there is no point spending anytime on them but ai will flag to you if an exceptional candidate comes along and you may want to bend some rules for it.

@WuTangAI @gabriel1 so spend 30 times the effort for both parties for what essentially boils down to the same thing 😭 quality candidates would also be filtered out that way too

Weird how we hate reading ai written text on web while the same text if generated by claude/ chatgpt in your own setup will feel mostly normal. The frame changes the absorption of knowledge,

Using @Remotion to customise learning for every individual is an underrated opportunity. Imagine a tool that first takes a small interview of yours asking similar questions and creates a map of your responses to understand what is the optimal way you learn a topic and then curate the learning path by generating visuals. This is so much important at school level, while each kid is unique but due to limited supply of teachers they are forced to tune learning in a traditional way with ai each individual should be taught with an Individual curriculum

Ai is tool or not totally depends on the user. Lot of people will become tool for ai and for a few it will act like the greatest tool ever created.