|
WTF, Daily
Wondering what the fuck is going on each
day? Same.
Tuesday, April 14, 2026
|
|
Good morning — oil has come back down from
its perch, Mark Zuckerberg has unveiled
something called Muse Spark, and the IMF has
found reasons to be both cheerful and
alarmed, which is more or less its natural
state. Here's what happened.
|
|
|
Meta Unveils Muse Spark and Quietly Buries
Its Open-Source Principles in an Unmarked
Grave
Meta has released Muse Spark, the inaugural
model from its grandly named
Superintelligence Labs, and the occasion is
notable for at least two reasons — the model
itself, which performs respectably across
multimodal, reasoning, and agentic tasks,
and the rather significant fact that it is
entirely closed-source. This represents a
pivot of some consequence for a company that
has spent the better part of four years
positioning itself as the principled
champion of open AI, waving the Llama models
about like a banner. The man responsible for
the change is Alexandr Wang, the Scale AI
co-founder whom Zuckerberg recruited as
chief AI officer for a consideration that
included $14.3 billion and a 49% stake in
Wang's former company — a sum that, one
supposes, buys a certain latitude to reverse
long-held positions.
Muse Spark, originally code-named Avocado —
a fruit that contains multitudes — is now
available in Meta AI on the web and the
company's app, with deployment to Facebook,
Instagram, WhatsApp, Messenger, and the
Ray-Ban Meta glasses to follow. The model's
arrival signals that Wang's nine months of
tenure have produced something Meta is
confident enough to show in public, which is
itself a minor milestone; the company's
recent AI efforts have had the consistency
of a soufflé prepared during a fire drill.
Whether Muse Spark represents genuine
superintelligence, or merely a very capable
conventional model dressed in aspirational
branding, is a question the benchmark
results will answer in approximately the
next forty-eight hours.
|
|
|
The AI Benchmark Arms Race Reaches Heights
That Would Alarm Anyone Who Understood What
They Meant
The latest rankings of frontier AI models
present a picture that would have been
dismissed as science fiction by anyone
reading it in 2022, and possibly by a few
people reading it now. Anthropic's Claude
Opus 4.6 and Google's Gemini 3.1 Pro are
currently leading the field with accuracy
scores exceeding fifty percent on Humanity's
Last Exam — a benchmark deliberately
constructed to resist AI capabilities by
drawing on problems that, as the name rather
heavily implies, were expected to be beyond
machine reach. Humanity's Last Exam was
launched as the benchmark that would hold
the line. It has been holding it for
approximately eight months.
The commercial picture is equally brisk:
Anthropic is approaching $19 billion in
annualised revenue, while OpenAI has
surpassed $25 billion and is making early
preparations for a late-2026 IPO — a
sequence of numbers that would have prompted
the average venture capitalist, circa 2019,
to lie down in a darkened room. The
generative AI tools available to American
consumers are now estimated to generate $172
billion in annual value, with the median
per-user figure having tripled between 2025
and 2026, suggesting that whatever is
happening with these models, at least some
of it is proving useful to people who are
not primarily interested in the benchmarks.
|
|
|
|
|
Trump Says Iran "Wants a Deal," and Markets,
Desperate for Good News, Choose to Believe
Him
Less than forty-eight hours after ordering a
naval blockade of the Strait of Hormuz and
sending oil past $103 a barrel, President
Trump has reversed the emotional trajectory
of global markets by announcing that Iranian
officials have telephoned and expressed a
desire to negotiate. "They want to work a
deal," Trump informed reporters with the air
of a man who has been expecting this call
all along, which may or may not be true but
is certainly effective messaging. Oil
retreated sharply on the news, European
shares advanced 0.6%, and Asian technology
stocks led a 1.8% regional rally — the
financial equivalent of a collective exhale
from an audience that had been holding its
breath since Sunday.
Pakistan has again offered to host the next
round of talks in Islamabad, displaying a
commitment to diplomatic hospitality that
deserves some form of recognition even if
the talks themselves continue to conclude in
the same manner as the previous ones. The
structural difficulties — Iran's insistence
on a ceasefire in Lebanon and the release of
frozen assets, America's demand for
enrichment to stop, the Strait to reopen,
and the general atmosphere of mutual
distrust — remain exactly as they were on
Saturday, which is to say formidable. The
markets, however, are not philosophers; they
have decided that a phone call constitutes
progress, and one hesitates to argue with a
1.8% rally before breakfast.
|
|
|
Pam Bondi Declines Her Congressional
Deposition, and Democrats Discover They Have
Nowhere to Send the Subpoena
The House panel investigating the Epstein
files has encountered a procedural wrinkle
of the kind that would delight a
constitutional law professor and infuriate
everyone else: Pam Bondi, the former
Attorney General who was subpoenaed in her
capacity as Attorney General, will not be
appearing for today's scheduled deposition
because she is no longer the Attorney
General. The Justice Department's position,
delivered with the sort of technical
precision that suggests several lawyers
thought about it carefully, is that the
subpoena was addressed to an office that
Bondi no longer holds, and one cannot,
apparently, compel a private citizen to
appear by simply addressing the paperwork to
a title they once held. The committee will
now contact Bondi's personal counsel "to
discuss next steps," a phrase that
translates, in practice, to "starting the
process over."
The Epstein probe has been generating this
sort of procedural friction with the
regularity of a mechanism specifically
engineered to produce it. Democrats, who
have pursued the matter with the persistence
of a terrier that has located something
interesting behind the skirting board,
continue to insist that the public interest
in the files outweighs the administrative
inconveniences. Meanwhile, California
Representative Eric Swalwell, who withdrew
from the state's governor's race on Sunday
night amid sexual misconduct allegations, is
now facing calls from fellow Democrats to
resign from Congress altogether — a reminder
that this particular Tuesday in Washington
is not running short of incident.
|
|
|
|
|
Scientists Discover 110 New Species in the
Coral Sea, Confirming the Ocean Has Been
Hiding Quite a Lot
CSIRO researchers, working aboard the vessel
RV Investigator in collaboration with the
Nippon Foundation-Nekton Ocean Census, have
announced the discovery of more than 110
species new to science in the deep waters of
the Coral Sea Marine Park — a number that
rather makes one feel that the planet has
not yet finished introducing itself. The
creatures were found at depths between 2,000
and 3,000 metres, in a region that has
historically received rather less scientific
attention than waters closer to the surface,
on the grounds that the crushing pressure
and total darkness make it an inconvenient
place to conduct fieldwork. Among the new
arrivals are four previously unknown fish
species — a chimaera, a deepwater catshark,
a stingaree ray, and a skate — plus a
remarkable quantity of invertebrates whose
formal descriptions are still being
compiled.
The headline figure sits within a larger
tally that underscores the ocean's talent
for concealment: in its first twelve years
of operation, RV Investigator has helped
describe 179 species new to science, yet
scientists estimate that more than 1,500
additional new species from collected
specimens are still awaiting formal
description — essentially waiting in a
queue, preserved in jars, while their
paperwork is processed. The smallest of the
new Coral Sea finds is a crustacean
measuring 2.5 millimetres, collected from
four kilometres beneath the surface, which
suggests that the ocean is not merely vast
but also committed to making its smallest
residents as difficult as possible to
notice.
|
|
|
Cancer Researchers Convene in San Diego,
Armed With Rather More Promising Data Than
Last Year
The American Association for Cancer Research
Annual Meeting opens Thursday in San Diego,
drawing researchers from across the oncology
field to present what promises to be a
particularly rich slate of data — rich
enough that the conference has been
described, in the measured language of press
releases, as "highly anticipated," which in
scientific circles is roughly equivalent to
standing ovations. Among the presentations
expected to generate the most discussion are
advances in personalised cancer vaccines,
which use mRNA technology to train the
immune system against a patient's specific
tumour mutations; updated results from
antibody-drug conjugate trials, which have
been delivering survival benefits that were,
until recently, considered optimistic; and
early data on AI-assisted diagnostics
capable of detecting cancers at stages where
treatment outcomes are substantially better.
The meeting arrives at a moment when the
broader oncology pipeline is, by historical
standards, unusually productive. Mortality
rates for several major cancer types have
declined in each of the past five years,
driven by a combination of earlier detection
and treatments that are increasingly precise
about which cells they destroy and which
they leave alone. Researchers at Penn
Medicine's Abramson Cancer Center will
present new findings on CAR-T cell therapies
and liquid biopsy applications — areas where
the gap between what is theoretically
possible and what is clinically available
has been narrowing with a speed that even
the optimists find gratifying. The
conference runs through April 22, after
which a substantial number of papers will be
published in journals where the rest of us
can read them and feel cautiously
encouraged.
|
|
|
|
|
The IMF Upgrades Its Global Growth Forecast,
Then Immediately Explains Why One Shouldn't
Get Too Excited
The International Monetary Fund has released
its April 2026 World Economic Outlook — a
document that has the feel of a school
report written by someone who genuinely
likes the student but has spent rather too
long in the staffroom. The headline number
is 3.3% global growth for 2026, an upgrade
from October's estimates that reflects, per
the IMF, genuine resilience in technology
investment, more supportive-than-expected
fiscal policy, and a private sector that has
adapted to post-pandemic trade disruptions
faster than the models anticipated. The
United States receives the organisation's
most positive major-economy assessment,
powered by AI investment, consumer spending,
and the tax provisions of the One Big
Beautiful Bill Act — a legislative
instrument whose name continues to generate
more commentary than its contents.
The caveats arrive promptly, in the manner
of a butler who has been standing just
outside the door. The upgrade comes with
warnings about inflation trajectories,
conflict risks, defence spending pressures,
and what the IMF calls "sharply less equal"
distribution of growth — the last being a
polite way of noting that 3.3% global growth
covers a great deal of ground between those
doing rather well and those doing rather
poorly. The Iran conflict represents the
organisation's most prominent near-term
risk, given its potential to extend energy
disruptions through a supply chain that was
already operating with less slack than
anyone would ideally prefer. The report is,
in short, the economic equivalent of a
doctor who delivers good news and then
suggests one sit down before the next part.
|
|
|
Goldman Sachs Reports $17 Billion in
Revenue, and Wall Street's Fee Machine Hums
Contentedly Along
Goldman Sachs opened earnings season Monday
with net revenues of $17.23 billion and net
earnings of $5.63 billion for the first
quarter — numbers that, read aloud in a room
containing people who were told in 2022 that
rising rates would end the investment
banking party, might produce a certain
awkward silence. Diluted earnings per share
came in at $17.55, comfortably ahead of
estimates, driven by investment banking fees
that surged as companies rushed to complete
deals before geopolitical uncertainty could
complicate the pipeline further. JPMorgan
reports this morning, with analysts
expecting earnings per share of $5.45 on
revenues of around $49 billion, which would
represent the sort of quarter that most
businesses would regard as a thoroughly
satisfactory year.
The broader earnings season — which also
brings Netflix, BlackRock, Johnson &
Johnson, and Morgan Stanley to the
confessional in the coming days — is being
watched as what analysts are calling the
"definitive litmus test" for the American
economy, a phrase that will seem prescient
or overwrought depending entirely on what
the numbers say. The particular curiosity
this quarter is how banks are characterising
the Iran conflict's effect on dealmaking
appetite: whether clients have paused,
pivoted, or — in the manner of experienced
capital allocators who have seen several
crises and learned to work around them —
simply carried on. Goldman's results
suggest, at minimum, that the last group
remains numerically significant.
|
|
WTF, Daily
The news, without the nonsense. Mostly.
|
|