reasonabl_human

Look into Nix / NixOS

What tests are you running? Anything outside of the alpaca eval test suite?

Also interested. Would be really great to run this against a self hosted LLM agent.

Does this mean your queries are all one-shot instead of utilizing techniques like LangChain?

‘Big Tech’ will always be there in some form down the road. If you have the risk tolerance and early career flexibility, take the path less traveled.

Black bar?

ControlNet is the new development that will really allow us to guide diffusion model outputs more granularly - first I’m seeing it used against video generation

Thanks for your work in this space, it’s incredible!

I’ve been sidetracked with work but planning on tuning llama 65B to produce alpaca 65B, will distribute via huggingface or torrent..

FWIW running the 30B alpaca-lora model quantized to 4-bit via llama.cpp has given me great results, and while I don’t expect much of an improvement from 65B at FP16, 65B will probably perform better than 30B when quantized

The interesting next steps in my head are more focused around curating a better instruction-tuning dataset using GPT-4, then fine-tuning again, and integrating the LangChain project with the resulting agent

What is the closest approach we know of today that plays games, not plays? The dialogue above is compelling, and makes me wonder if the same critique can be levied against most prior art in machine learning applied against games. E.g. would you say the same things about AlphaZero?

Okay I’ll bite - please elaborate!

Interesting, where can I learn more about prompting, and tuning a raw model?

Thanks so much for sharing these insights! Especially the humanizing points around life still unfolding around you regardless of how much success you achieve.

Interesting- in that case these are some eye popping numbers.

However, I’m struggling to square that fact with the observation that the Wikipedia table shows a number of entries >100%. How could government spending ever exceed 100% of GDP given it is counting toward GDP?

Can someone explain the significance of this statistic in context? Seems there are many successfully economies with a government spend of higher % of GDP than the US. Not sure how this fits into a broader picture of a country’s financial health.

Also- based on my cursory reading here, it seems disingenuous to claim 46% of GDP is government spending- the stat is that government spending is comparable to 46% of the GDP of that year. Government spend doesn’t comprise 46% of the GDP, rather we are using GDP as a measuring stick to evaluate the sanity of government spending amounts.

Whoa! Do you have a build log posted somewhere? What platform did you build on?

I think I just discovered a dangerous new rabbit hole…

Do you have recommended resources for getting into #3, specifically around funding / assisting others in building businesses?

I am interested in doing so outside of the traditional VC models and tech sector, but the signal to noise ratio of information on building businesses is incredibly low on the web.

Where do you folks go to learn on these macroeconomic topics?

Wow. Really grounding to see the personal troubles of such a monumental figure.

What an inspiring lecture… thanks for posting!

I seriously cannot tell if this is satire

I’ve always wondered how this works- sure these folks are already personally wealthy, and they have large stock portfolios, but the mechanics of tapping into that for liquidity is interesting to me..

Say an exec makes Jassy’s base and has a large equity packages. Do banks feel comfortable writing loans against to-be-vested RSUs? I’d imagine they count RSU vesting as heavily discounted future cash flow.. but not sure.

Plus there is significant volatility in the actual value they bank at the end of the day- sure, speaking on normal terms having an equity package fluctuate at this scale isn’t significant to an individual’s finances, but might be to a bank underwriting, say, a recent Snap exec’s loans…

Finally someone sane. It’s not about stacking 400x boxes- it’s about force multiplication and impact

Is this something you use personally?

Wow. What kind of tech debt are you reckoning with / what better design decisions would you implement if redesigning today?

Mind blowing.. dang!!

Yeah this is comical. State’s rights have eroded so much since the inception of the country that it’s considered strange or experimental when a state governs with autonomy as designed.

There is no way these were internal recruiters…. third-party head hunters?

Fun delivery on a dense topic

Couldn’t stand the small rubber nub function keys, and the as you mentioned the echo was off-putting. Also zero customizability in the split nature since it’s one massive plastic hunk…. Excited for the advantage 360 this summer as they finally are addressing this by going split and ditching the function keys