Jeffrey Salemi

Craft beer, Boston sports, video games, and pediatric medicine. What a combo.

[Jeffrey Salemi @DeliMeatMD]

@jumperz 1) wtf is the Y-axis scale? just arbitrary increasing time units? 1min to 10min x10, 10min to 1-hour x6, 1 hour to 8 hour x8 but lets keep it all evenly spaced. 2)50% reliability is embarassingly bad as a bar for any task. If an employee was 50% reliable they will be fired

[Jeffrey Salemi @DeliMeatMD]

@unusual_whales the mechanism is called bankruptcy btw

[Jeffrey Salemi @DeliMeatMD]

@ahall_research yeah guys! slowdown before we cure cancer! were definitely so close we have to slow down! let me IPO first!

[Jeffrey Salemi @DeliMeatMD]

@a16z What's even less predetermined is that we create a technology where these conversations even fucking matter. "We don't have to end up dystopian" Yeah. We could just end up with expensive autocomplete too given diminishing returns on scaling laws 🙄

[Jeffrey Salemi @DeliMeatMD]

@gbrew24 oh an axios link?!? it must be true!!!!!

[Jeffrey Salemi @DeliMeatMD]

@Rory_Johnston Feels like the market is pricing in global recession without the oil price spike that would actually drive demand destruction/growth deceleration.

[Jeffrey Salemi @DeliMeatMD]

If LLMs are becoming recursive and approaching AGI, why bother releasing consumer or even business facing products which are profit negative? Why not go all in on the recursive LLM products and spring full speed toward AGI?

[Jeffrey Salemi @DeliMeatMD]

@TheStalwart The framing of this question implies a linear/exponential trajectory of LLM improvements with an endpoint of essentially AGI. The jury is still out on how capable LLMs can even become as they are now having diminishing model returns and requiring neurosymbolic inputs.

[Jeffrey Salemi @DeliMeatMD]

@gbrew24 @gzeromedia They have been "close to a deal" for 10-weeks. This kind of reporting is just disingenuous at this point.

[Jeffrey Salemi @DeliMeatMD]

@cory_schulz_ @Hesamation Given the integration of neurosymbolic architecture and tools (harnesses etc), I see it as an indictment of the pure LLMs capacity. Mythos is also going to be sub 100%

[Jeffrey Salemi @DeliMeatMD]

@bridgemindai Given cost estimates, no. Now that costs are token based the more competent a model is, the more expensive it is to run it. The key ratio is 8x code to what number of tokens used and compare that to prior models

[Jeffrey Salemi @DeliMeatMD]

Gonna start using "its early" in my everday vernacular like AI boosters. "My sandwich has turned into solid gold yet, but its early"

[Jeffrey Salemi @DeliMeatMD]

Can someone please explain to me like I'm 5yo how "the next frontier is physical AI" when LLMs can't successfully tell me how many Rs are in strawberry?

[Jeffrey Salemi @DeliMeatMD]

@Yuchenj_UW Option 3 only works if you presume the cost of unlimited tokens is lower than the cost of 50% of your labor force. Likely they will layoff half and still pay more

[Jeffrey Salemi @DeliMeatMD]

@gemoscatelli The only explanation to a recent 1-3mo trend is that their population is smaller than estimated? do you know how long a human lifespan is?

[Jeffrey Salemi @DeliMeatMD]

@StanWongWealth EV adoption is hilarious. Yes in the last 12-weeks drivers began dropping $50,000 for EVs in response to $4 gas.

[Jeffrey Salemi @DeliMeatMD]

@dtarian04 is your july WTI target still $75?

[Jeffrey Salemi @DeliMeatMD]

@dtarian04 This is aging like a fine wine and I’m here for it

[Jeffrey Salemi @DeliMeatMD]

@B_radRiffs @CNBC Good advice