Tips for AI writers:

1. Spend 30% of your effort on skimming all student ML papers (e.g. Stanford NLP CS224n) the past 3 years and prototype your favorites

The idea is everything. Pick an area you are interested in and ideally something that has a visual aspect to it

*Heads-up that this thread will feature some extreme examples of homophobia and transphobia.*

Hey folks, have you ever wondered why trans people face constant accusations of fetishism, sexual predation and child abuse/grooming? Well, today let's talk about stochastic terrorism.

So, first we have to talk about what stochastic terrorism actually is. Remember when there was a conspiracy that there was a pedophile ring operating out of a pizza parlour basement, and a crazed gunman showed up? It's that sort of incitement that stochastic terrorism describes.

It's the demonisation or incitement against a group of people or individuals with the intent that other, unaffiliated parties will act upon it; it's the releasing doctored footage and a shooter show up at Planned Parenthood as a direct result of what he's been led to believe.

So with that in mind, let's talk about Russia's extreme anti-LGBT laws that banned the "promotion" of LGBT lifestyles to children. A move that specifically marked LGBT people as a threat to children, and resulted in neo-nazis meting out vigilante "justice" against gay men.

Groups such as this would lure unsuspecting gay men via dating sites, and brutalise and humiliate them on camera. They did so under the guise of combating child abusers, one group calling themselves "Occupy

Latest update from @TheBernician

1/19

Hancock, Whitty, Vallance & Ferguson Stand Accused of Pandemic Fraud.

After yet more damning evidence arose last week, further additions had to be made to the court papers, which we are now aiming to lay in the coming week.

2/19

The four defendants will be Matt Hancock, Chris Whitty, Patrick Vallance and Neil Ferguson, each of whom has inextricable and ultimately incriminating connections with Bill Gates and Big Pharma.

3/19

As well as three counts of fraud by false representation and seven counts of fraud by non-disclosure, under sections 2 and 3 of the Fraud Act 2006, we can now also prove that the entire worldwide scamdemic originated on these shores.

4/19

In fact, we can show that, without the dishonest statements and non-disclosures of the four defendants we are initially proceeding against, COVID-1984 would not have transpired.

5/19

Furthermore, the 1st defendant, Matt Hancock, cannot rely upon the defence of Parliamentary Privilege, on the ground that none of the evidence relied upon by the prosecution is capable of preventing parliamentarians from carrying out their lawful business in Parliament.

1/ Throughout my startup career, I’ve seen many founders have an aversion to sales. This is a mistake that can kill your startup.

2/ Sales is often viewed as either a saving grace or proof that the product isn’t good enough (because it should sell itself). Neither are ever true. Some common mistakes that result in...

3/ Mistake 1: Hire a sales rep before reaching product/market fit to get your initial batch of customers. This is a mistake because founders need to work through their MVP with early adopters to truly understand what it is they’re selling.

4/ Mistake 2: Reach product/market fit, need to scale, and rely entirely on self-serve. For enterprise products that require big commitments and internal shifts, almost no product is self-explanatory enough to sell itself.

5/ Mistake 3: Make a first sales hire who isn’t scrappy enough to help mold the sales process from scratch. Some salespeople are amazing at their jobs, but not cut out to establish the processes that others end up following. This skillset is what @rdedatta calls a “sales ninja”.

Important paper from Google on large batch optimization. They do impressively careful experiments measuring # iterations needed to achieve target validation error at various batch sizes. The main "surprise" is the lack of surprises. [thread]

https://t.co/7QIx5CFdfJ

The paper is a good example of lots of elements of good experimental design. They validate their metric by showing lots of variants give consistent results. They tune hyperparamters separately for each condition, check that optimum isn't at the endpoints, and measure sensitivity.

They have separate experiments where the hold fixed # iterations and # epochs, which (as they explain) measure very different things. They avoid confounds, such as batch norm's artificial dependence between batch size and regularization strength.

When the experiments are done carefully enough, the results are remarkably consistent between different datasets and architectures. Qualitatively, MNIST behaves just like ImageNet.

Importantly, they don't find any evidence for a "sharp/flat optima" effect whereby better optimization leads to worse final results. They have a good discussion of experimental artifacts/confounds in past papers where such effects were reported.